SAS Viya MCP Server

A Model Context Protocol (MCP) server for executing SAS code, training AutoML projects, scoring models and so much more for SAS Viya environments.

Features

• 92 tools across 10 selectable tiers, spanning the Analytics Life Cycle on SAS Viya
• Interactive views (MCP Apps) for data, SAS logs and glossary editing in clients that render them — Claude, ChatGPT, Microsoft 365 Copilot, VS Code, Cursor
• Prompt Templates for improving your SAS Code
• OAuth2 authentication with PKCE flow
• HTTP-based MCP server compatible with MCP clients

Articles & Videos

Here you can find getting articles on how to use and integrate the SAS MCP Server in different tools and what to build with it:

• From REST APIs to AI Agents: Why the SAS Viya MCP Server Matters
• Connecting GitHub Copilot to SAS Viya with the SAS Viya MCP Server
• Bring Your Own Key: SAS Viya MCP Server with GitHub Copilot CLI
• Putting the SAS Viya MCP Server to Work in GitHub Copilot
• Connecting Claude Code CLI to SAS Viya with the SAS Viya MCP Server
• Putting the SAS Viya MCP Server to Work in Claude Code CLI
• Integration with SAS Retrieval Agent Manager (RAM)

Getting Started

Prerequisites

• Required

  • Python 3.12+
  • uv 0.8+
  • SAS Viya environment with compute service
  • Setup the Viya environment for MCP
    • See configuration.md

• Optional

  • Docker: refer to container setup
  • Kubernetes: sample manifests (Contour or nginx) and a Helm chart in deploy/

Installation

1. Clone the repository:

git clone <repository-url>
cd sas-mcp-server

2. Install dependencies

uv sync

NOTE: This will by default create a virtual environment called .venv in the project's root directory.

If for some reason the virtual environment is not created, please run uv venv and then re-run uv sync.

Usage

1. Configure environment variables:

cp .env.sample .env

Edit .env and set

VIYA_ENDPOINT=https://your-viya-server.com

2. Start the MCP server (see Choosing a deployment mode below):

Option A: HTTP mode (pre-run the server, connect from MCP client)

uv run app

The server will be available at http://localhost:8134/mcp by default. Authentication is handled via OAuth2 PKCE flow in the browser.

Option B: Stdio mode (MCP client starts the server on demand)

Authenticate once. Two equivalent options:

# Option B1 — if you have the SAS Viya CLI installed:
sas-viya auth loginCode

# Option B2 — built-in helper, no external CLI needed (Viya 2022.11+):
uv run sas-mcp-login

Both flows write an access token to a local cache (~/.sas/credentials.json and ~/.sas-mcp-server/credentials.json respectively); the stdio server reads whichever it finds. When the token expires, re-run the same command.

Then configure your MCP client to launch the server directly (see below).

Option C: Docker / Podman (containerized deployment)

Pull the pre-built image from GitHub Container Registry:

docker pull ghcr.io/sassoftware/sas-mcp-server:latest
docker run -e VIYA_ENDPOINT=https://your-viya-server.com -p 8134:8134 ghcr.io/sassoftware/sas-mcp-server:latest

Or build locally from source:

docker build -t sas-mcp-server .
docker run -e VIYA_ENDPOINT=https://your-viya-server.com -p 8134:8134 sas-mcp-server

Available image tags:

• latest — most recent tagged release
• <major>.<minor>.<patch> (e.g. 1.0.0) — specific release
• <major>.<minor> (e.g. 1.0) — latest patch of a minor release
• edge — tip of main (unreleased, for testing)
• sha-<short> — pinned to a specific commit

Programmatic clients with a pre-existing Viya token

If your caller already holds a Viya access token (e.g. an automation script that obtained one via the SAS Viya CLI), start the HTTP-mode server with ALLOW_RAW_BEARER=true and pass the token directly:

curl -H "Authorization: Bearer $VIYA_TOKEN" http://localhost:8134/mcp ...

The server validates the token against Viya's JWKS and uses it upstream as-is, bypassing the MCP JWT swap. The default OAuth2 PKCE flow keeps working alongside — both client types share the same /mcp endpoint.

If your Viya APIs are intentionally exposed without auth (for example, a local/dev Compute API endpoint), set VIYA_AUTH=false to bypass all SASLogon/OAuth flows in both HTTP and stdio modes. In this mode the server sends upstream requests without an Authorization header.

If your compute deployment does not expose /compute/contexts and only supports a fixed session, set COMPUTE_SESSION_ID=<session_id>. The compute tools will use that session directly instead of creating context-backed sessions.

Choosing a deployment mode

Quick guidance:

• Starting out or exploring? Use stdio — one sas-viya auth loginCode or uv run sas-mcp-login, then your MCP client manages the server lifecycle.
• Need secure, interactive auth? Use HTTP — no stored passwords, each user authenticates via browser.
• Deploying for a team or on a server? Use Docker — portable, no Python dependency on the host, easy to integrate with orchestrators.
• Running it for a whole organisation? Use Kubernetes — sample manifests and a Helm chart are in deploy/, including the routing the OAuth flow needs for either Contour (the chart's default, and the only one that can mount the server under a path prefix on an existing hostname) or ingress-nginx.
• Using Gemini CLI? Use stdio — Gemini CLI does not support HTTP mode or browser-based OAuth. See Gemini CLI configuration.
• Installing from a client's server catalogue? That path runs the published container in stdio mode (app-stdio), not as an HTTP server, so it authenticates from your ~/.sas token cache — which has to be mounted into the container at /app/.sas.

Limiting exposed tools (tiers)

Tools are grouped into numbered tiers. By default the server exposes all of them; set MCP_TIERS to expose only a subset — handy for keeping a client's tool list small and focused, or hiding capabilities a deployment shouldn't offer. Accepts ranges and comma lists (e.g. MCP_TIERS=0-4 or MCP_TIERS=0,1,6,7); unset means all tiers.

# Example: expose only compute/discovery/data-ops and reporting
MCP_TIERS=0-3 uv run app

Read-only mode

Set MCP_READ_ONLY=true to expose only tools that neither change server-side state nor cause server-side work — 51 of the 92 tools. Withheld tools are never registered, so they are absent from the client's tool list entirely: the model cannot see them, so it cannot attempt them.

This is a filter over the tiers, not a tier of its own — the read/write split cuts across every tier (Tier 3 has both get_report and delete_report). The two settings compose:

# Every read tool, all tiers
MCP_READ_ONLY=true uv run app

# Read tools of the reporting and decisioning tiers only
MCP_TIERS=3,7 MCP_READ_ONLY=true uv run app

The definition is strict: a tool qualifies only if it can neither write nor start work. Beyond the obvious create/update/delete tools, that withholds:

Classification is fail-closed: a tool that is not explicitly classified as read-only is withheld. The list lives in src/sas_mcp_server/tools/_access.py, and a test asserts it covers every registered tool, so a newly added tool cannot silently land in read-only mode.

Tool annotations (what clients are told)

The same classification is advertised to every client as MCP tool annotations on each tools/list entry — whether or not read-only mode is on:

Clients use these to shape their approval UX — e.g. Claude groups read-only tools for one-click approval and warns before destructive ones — and to decide when to interrupt the user. They are hints, not enforcement: the spec tells clients to treat them as untrusted unless the server is trusted, and MCP_READ_ONLY remains the server-side control. Without annotations a client must assume the spec's pessimistic defaults (writable, destructive, open-world) for every tool, so this only ever reduces friction. The browser landing page marks each tool read-only / write / destructive from the same hints.

Interactive views (MCP Apps)

Some results read badly as text however well the tool shapes them: a hundred rows of a table, a SAS log, a form's worth of typed attributes. Ten tools therefore also carry a small HTML view, using the MCP Apps extension, that a supporting client renders in a sandboxed frame beside the result:

Every view has a ⤢ button that opens it fullscreen where the host offers that (Claude does), which is where a wide table or a long log is meant to be read; the button does not appear on a host without it.

Everything about this is additive. The tool's return value is unchanged, so a client without the extension — Claude Code, a script, an older client — sees exactly what it saw before. Clients that render views today include Claude (web, desktop and mobile), ChatGPT, Microsoft 365 Copilot, VS Code with GitHub Copilot, Cursor and Goose. Every request a view makes travels through the host to this server as an ordinary tool call with the session's token; the browser never talks to Viya, and a tool a deployment withholds through MCP_TIERS or MCP_READ_ONLY has no view either — nor does any control inside a view that would have called it, so a read-only deployment shows the glossary browser without its Publish button rather than with one that can only fail. The pages are fully inlined — the vendored @modelcontextprotocol/ext-apps bridge included — so they render on hosts that ignore CSP declarations and on an air-gapped Viya. MCP_APPS=false registers no views.

Available Tools

The headings below match the numbered tiers above, so MCP_TIERS maps directly to the tools you expose (e.g. MCP_TIERS=0-3 gives Tiers 0–3).

Tier 0 — Compute Contexts & Code Execution

• execute_sas_code: Execute SAS code snippets and retrieve execution results (log and listing output). Runs in a reusable, per-user compute session that is kept warm across calls, so SAS state (WORK tables, macro variables, assigned librefs) persists between successive calls — use reset_compute_session to start fresh.
• list_compute_contexts: List available compute contexts
• reset_compute_session: Delete the cached compute session for a context, discarding its SAS state and forcing a fresh session on the next call

Tier 1 — Data Discovery

Information Catalog (metadata discovery & profiling):

• catalog_search: Search the catalog for assets (tables, columns, reports, …) using the SAS catalog search grammar (free text, facets like AssetType:Report, ranges). Each hit carries a resource_uri you can hand to the matching tool (e.g. get_report, get_castable_data).
• catalog_search_helper: Discover how to query the catalog — list the available facets, or the valid values for one facet — so you can build precise catalog_search queries.
• catalog_find_instance: Resolve the catalog instance for a source-asset resource_uri, bridging a search hit to the profiling and download tools without handling an instance id by hand.
• catalog_run_adhoc_analysis: Submit an ad-hoc profiling job for a table. NLP enrichment (language, sentiment, semantic IDs) is on by default, populating informationPrivacy, nlpTerms, nlpTags, and mostImportantFields.
• catalog_get_adhoc_analysis: Poll a profiling job and cross-check the target instance, reporting profile_ready once results have landed on the asset — so a download isn't fired too early.
• catalog_download_table_profile: Download a table's data dictionary and column profile as CSV, identified by either instance_id or resource_uri.
• catalog_list_agents: List the catalog's discovery agents (the crawlers that populate metadata).
• catalog_run_agent: Start a discovery agent run (asynchronous) to crawl its data source and refresh catalog metadata.
• catalog_get_agent_history: Inspect an agent's run history — status and how much metadata each run enumerated/added/updated/removed.

CAS data (in-memory):

• list_cas_servers: List available CAS servers
• list_caslibs: List CAS libraries on a server
• list_castables: List tables in a CAS library
• list_source_tables: List source tables not yet loaded into memory (candidates for promotion)
• get_castable_info: Get table metadata (row count, columns, size)
• get_castable_columns: Get column names, types, labels, formats
• get_castable_data: Fetch sample rows from a CAS table
• query_data: Run a FedSQL SELECT against CAS or compute data and get the rows back — one SQL surface over both storage tiers. Pick the tier with target (cas for caslib.table, compute for libref.table); joins, subqueries, aggregation, and UNION all work, and the row cap is applied server-side by the tool (a LIMIT you write is ignored, since a malformed one is silently discarded by CAS). Optionally returns the query as CREATE VIEW text for you to run yourself. Reads only: writes are refused pre-flight, and SAS macro triggers (%/&) are rejected because the macro processor would expand them outside SQL. Note the two tiers cannot be joined in one statement.

Compute libraries (SAS/Compute, within a compute context):

• list_compute_libraries: List the SAS libraries (librefs) assigned in a compute context
• list_compute_tables: List the tables in a SAS library within a compute context
• list_compute_columns: List the columns of a table in a SAS library
• get_compute_table_data: Fetch a page of rows from a table in a SAS library, values formatted as SAS displays them — the compute-tier counterpart of get_castable_data

Tier 2 — Data Operations & Files

• upload_data: Upload a data file into a CAS table — read server-side so the data never passes through the model's context — from file_path (the server reads it off disk) or url (the server fetches it and converts it to the multipart upload the endpoint requires). Ingests the formats the casManagement uploadTable API accepts — csv, tsv (csv + tab delimiter), xls, xlsx (single sheet), sas7bdat, sashdat — auto-detected from the extension or set with data_format. parquet is not accepted by that endpoint and is rejected up front with guidance (load via a path-based caslib + promote_table_to_memory, or convert to csv/sas7bdat).
• upload_inline_data: Create a small CAS table from inline csv/tsv text passed as a string (a lookup/mapping table the model builds on the fly, or a quick test table). The payload travels through the model's context, so it's for tiny tables only — use upload_data for files or anything larger.
• promote_table_to_memory: Load a source table into memory at global scope (idempotent)
• list_files: List files in the Viya Files Service
• upload_file: Upload a file to the Viya Files Service, optionally into a Content folder (parent_folder_uri). Content comes from exactly one of content (inline text), file_path (read server-side, binary-safe — xlsx, zip, images — gated by ALLOW_LOCAL_FILE_UPLOAD), or url (server-side fetch)
• download_file: Download file content

Tier 3 — Reports & Visualization

• list_reports: List Visual Analytics reports
• get_report: Get report metadata and definition
• export_report: export a report (or specific report objects) in any format the VA service supports — package (zip), pdf, png, svg, csv, tsv, xlsx, or summary. Text formats come back inline, png as image content, and binary formats (package/pdf/xlsx) as an embedded file with the right MIME type.
• describe_report_objects: Discover what a report can contain — the eight report operations and every addable object (bar chart, list table, geo map, key value, …) with a one-line purpose, its data roles, common options, and an example payload. Call with no arguments for the catalog (including an intent→object map, placement guide, layout recipes, and the API's hard limits), object_type= for one object's contract (colloquial aliases like kpi resolve), category= to filter, or operation= for one operation's full shape — operation="addData" documents dataItems (column renames, SAS formats, aggregations, geography classification). Backs the apply_report_operations loop.
• create_report: Create a Visual Analytics report and return its id. Optionally pass an operations array to build the whole report in one atomic call; the result carries the created page/object names+labels and a verify hint.
• apply_report_operations: The authoring workhorse — apply an ordered batch of native VA operations (addData, addPage, addObject, updateObject, setParameterValue, updateData, changeData, applyDataView) to a report. Give a page a title with addPage's title field (a text band at the top of the page body — VA headers are controls-only); title every chart at add time via options.object.title; arrange objects with placement — page, relativeToObject (left/right/top/bottom for columns, rows, and grids), container (group into a standardContainer), or report (new_page creates-and-names a page inline for one-batch multi-page reports). The batch is atomic. Validates every operation, object key, and placement against the catalog first (reporting all errors at once), supports dry_run, handles the ETag concurrency handshake, and — with result_report_name/result_folder — applies the batch save-as to a new report, leaving the source untouched. Typical loop: describe_report_objects → get_castable_columns → apply_report_operations → get_report_outline / export_report (png, page-by-page) to verify.
• get_report_outline: Read a report's structure back — pages → objects with the handles the other tools need (object name for placement/updateObject targets, label for export_report, page label for page placement).
• copy_report: Copy a report to a new one (optionally renaming/refoldering). Pairs with a changeData operation for the copy-and-replace pattern.
• delete_report: Delete a report and its content.

Tier 4 — Batch Jobs & Async Execution

• submit_batch_job: Submit a SAS job for async execution
• get_job_status: Check job state
• list_jobs: List recent/running jobs
• cancel_job: Cancel a running job
• get_job_log: Retrieve job log

Tier 5 — Automated Machine Learning

• list_ml_projects: List AutoML projects
• create_ml_project: Create a new AutoML project from a loaded, global-scope CAS table (caslib + table + optional CAS server)
• run_ml_project: Run pipeline automation
• register_ml_champion_model: Register an AutoML project's champion model to the Model Repository
• publish_ml_champion_model: Publish an AutoML project's champion model to a scoring destination

Tier 6 — Model Management & Scoring

• list_registered_models: List models in repository
• list_publishing_destinations: List available scoring/publishing destinations, for use with publish_ml_champion_model
• list_mas_modules: List published MAS modules
• get_mas_module_step_signature: Inspect a published MAS module step's input/output variable signature before scoring
• score_data: Score data against a published model or decision

Tier 7 — Decisioning (SAS Intelligent Decisioning)

Build and manage SAS Intelligent Decisioning rule sets and decision flows end to end, then publish a flow to Micro Analytic Score (MAS) so score_data can execute it.

Business rules — rule sets:

• create_business_ruleset / update_business_ruleset / get_business_ruleset / list_business_rulesets / delete_business_ruleset: Manage rule sets (the input/output signature the rules operate on)
• lock_business_ruleset_revision: Lock the current rule set state as an immutable revision (what a decision step references)
• list_business_ruleset_revisions: List a rule set's locked revisions

Business rules — rules:

• create_business_rule / update_business_rule / get_business_rule / list_business_rules / delete_business_rule: Manage the conditional rules inside a rule set

Decision flows:

• create_decision_flow / update_decision_flow / get_decision_flow / list_decision_flows / delete_decision_flow: Manage decision flows that chain rule set steps
• get_decision_flow_code: Retrieve the generated DS2 execution code for a flow
• lock_decision_flow_revision / list_decision_flow_revisions / get_decision_flow_revision: Lock, list, and fetch immutable decision revisions
• publish_decision_flow: Publish a locked decision revision to a MAS destination, polling to completion and returning the server-generated MAS moduleId (directly usable with get_mas_module_step_signature / score_data)

Tier 8 — Workbench (Execute Code Only)

• execute_sas_code: Execute SAS code snippets and retrieve execution results (log and listing output). Runs in a reusable compute session that is kept warm across calls, so SAS state (WORK tables, macro variables, assigned librefs) persists between successive calls

Tier 9 — Business Glossary (SAS Data Governance)

Read and author the SAS Business Glossary, and link its terms to the columns they describe. Tier 1 tells you a column is called CD_NAC_RSK; this tier tells you what that means and who says so.

Two things about the glossary are worth knowing before you start, because both are invisible in the raw API and both are handled for you here:

• A term has two ids. It exists as a Glossary object and as a Catalog entity, with different identifiers. Every tool returns both — term_id (glossary) and catalog_entity_id (catalog) — so you never have to work out which one you are holding.
• Custom attributes are stored under UUID keys. These tools read and write them by the label the glossary UI shows ({"Scope": "Group"}), validating required attributes and single-select values before the call is made.

Dictionary:

• search_glossary_terms: Free-text, ranked search over term names and definitions — the way in when you know a word rather than an id. Reports assigned_asset_count, so you can see whether a term is actually in use
• list_glossary_terms: Exact structural listing — by term type, by parent (the authoritative hierarchy), or by name fragment. include_attributes returns each term's attribute values (free — the listing already carries them), and attribute_filter keeps only the terms matching, e.g. {"Used in Risk": true}. The glossary cannot filter on attributes server-side, so that filter is applied here and the result reports how much of the dictionary it scanned
• get_glossary_term: One term in full, with its custom attributes named rather than hashed
• list_glossary_term_types / get_glossary_term_type: The term types available, and the attribute contract a term of that type must satisfy — call the latter before authoring. Each attribute reports the one value format Viya accepts for it, which the API itself documents nowhere

Where terms meet data:

• list_term_assets: The columns a term is attached to, with their tables. The authoritative answer to "where is this term used?"
• list_table_terms: The reverse — every column of a table and the term assigned to it, with the term's definition inline. The fastest read on whether a table is governed

Authoring:

• create_glossary_term: Create a term. Publishes by default — the underlying API creates an invisible draft unless told otherwise; update_glossary_term(publish=true) promotes one later
• import_glossary_terms: Create many terms, and their hierarchy, in one call. Children name their parent instead of needing its id, so rows can be given in any order and no id is threaded between levels; per-row failures are reported individually. Returns the term_id of each row's term, so the next step needs no lookup, and separates rows that were genuinely new from rows whose term already existed — the import job counts both as successful. A row is written whole, so update_existing replaces the term at that path rather than merging into it. dry_run=true shows the resolved rows without importing — the interactive view turns that into a table with an Import button
• update_glossary_term: Change a term's text, parent or attributes. Merges onto the current term, so omitted fields are left alone rather than blanked; parent_id moves it in the hierarchy, and publish promotes a draft. A draft is a separate resource in the API, so this routes the write accordingly — editing one otherwise fails with a bare 404
• delete_glossary_term: Permanently delete a term and every assignment that referenced it. There is no cascade — a term with children is refused, so delete a subtree leaf-first
• assign_glossary_term / unassign_glossary_term: Attach a term to a table column, or detach it. This is the step that makes a term govern data — a term with no assigned assets governs nothing

Designing the vocabulary:

• create_glossary_term_type / update_glossary_term_type / delete_glossary_term_type: Define the template terms are created from — which custom attributes they carry, which are mandatory, and what values each accepts. An edit matches attributes by label and keeps each one's identifier, so terms already carrying a value do not lose it — to rename one, give its attribute_id alongside the new label, since a new label matches nothing and would otherwise mint a new attribute; delete refuses while terms still use the type

Terms assigned this way also become searchable through Tier 1's catalog_search using the Column.term:"<term name>" facet on the datasets index, which returns the tables carrying a term without resolving individual columns.

Prompt Templates

• debug_sas_log: Analyze SAS log for errors with root-cause explanations
• explore_dataset: Generate data-profiling SAS code
• data_quality_check: Generate DQ assessment code
• statistical_analysis: Set up a statistical workflow with diagnostics
• optimize_sas_code: Review and optimize SAS code
• explain_sas_code: Block-by-block code explanation
• sas_macro_builder: Build production-quality SAS macros
• generate_report: Generate ODS/PROC REPORT code
• build_va_dashboard: Guide a polished multi-page Visual Analytics dashboard build from a CAS table — a discover → shape → structure → polish → verify method over the report-authoring tools

MCP Client Configuration

Example configurations are provided in the examples/ folder. Below are quick-start snippets for common clients.

Tip — open the endpoint in a browser. In HTTP mode, pointing a browser at the MCP URL (e.g. http://localhost:8134/mcp, or https://<host>/mcp for a deployed server) shows a landing page instead of a bare 401: what the server is, which SAS Viya it talks to, the tool tiers this deployment exposes with a one-line summary per tool, and ready-to-copy configuration for Claude Code, VS Code, Cursor, Claude connectors and generic mcp.json clients — with the deployment's real URL already filled in. Only a plain browser GET (Accept: text/html) is answered this way; MCP clients and curl see exactly what they saw before. The page is unauthenticated and shows deployment shape only (never user data); administrators can turn it off with MCP_LANDING_PAGE=false.

VS Code / Cursor / Claude Code (.vscode/mcp.json)

HTTP mode (requires uv run app running separately):

{
    "servers": {
        "sas-execution-mcp": {
            "url": "http://localhost:8134/mcp",
            "type": "http"
        }
    }
}

Stdio mode (starts the server on demand):

{
    "servers": {
        "sas-execution-mcp": {
            "command": "uv",
            "args": ["run", "app-stdio"],
            "cwd": "${workspaceFolder}"
        }
    }
}

Gemini CLI (.gemini/settings.json)

Gemini CLI only supports stdio mode. Add to your ~/.gemini/settings.json or project-level .gemini/settings.json:

{
    "mcpServers": {
        "sas-viya-mcp": {
            "command": "uv",
            "args": ["run", "app-stdio"],
            "cwd": "/path/to/sas-mcp-server",
            "timeout": 60000
        }
    }
}

Note: The timeout field (in milliseconds) is important — SAS Viya API calls can take longer than the Gemini CLI default of 10 seconds. A value of 60000 (60s) is recommended. Set cwd to the absolute path of your sas-mcp-server checkout.

Example

Execute SAS code through the MCP tool:

data work.students;
input Name $ Age Grade $;
datalines;
Alice 20 A
Bob 22 B
;
run;

proc print data=work.students;
run;

For more details, configuration options, and deployment options, please refer to the examples folder and follow the instructions listed there.

Collection Mode (Usage Telemetry)

An opt-in, off-by-default mode that records how the server is actually used — which tools, for what goals, with what inputs, and where they fall short. It serves two audiences:

• Contributors giving structured feedback to the maintainers. Rather than filing prose bug reports, you can turn it on for a while and share the resulting log so maintainers can see which tools are used, which fail, and what goals have no good tool yet — a direct signal for improving existing tools and identifying new ones.
• Organizations running the server for their own users. Teams that deploy the MCP server internally can enable it to understand what their users do with it and why, entirely within their own infrastructure.

It is implemented as a FastMCP middleware wrapper (telemetry.py + usage_logger.py) and requires no changes to any tool.

🔒 Nothing is ever sent anywhere automatically. Collection mode only appends to a local log file on the machine running the server. It is disabled unless you explicitly enable it, and even when enabled the data stays on your disk — sharing it with anyone (including the maintainers) is a deliberate, manual step you take by sending the file yourself. There is no phone-home, no network transmission, and no third party involved.

When enabled it does two things:

1. Injects a required goal parameter into every tool's schema, asking the model to state in one sentence why it chose that tool for the current request. The goal is stripped from the arguments before the real tool runs, so tools never see it.

2. Appends one JSON line per tool call (JSON Lines / NDJSON, schema v3) to a local log file: timestamp, run id, per-run sequence number, tool name, goal, arguments (plus a stable args_hash for retry analysis), result, status, error, latency, and the calling client's client_name / client_version. When a tool declares a failure as data (e.g. {"status": "apply_failed"}, which the MCP layer sees as success), the record also carries tool_status / is_tool_error / tool_message / failed_operation_index — so tool-level failure rates are analyzable in every mode. A run_start header record (transport, pid, server version, result mode, and an optional COLLECTION_RUN_TAG label for tagging A/B runs) opens the log and is re-emitted every 1000 records, so rotation cannot leave a stretch of the log with no header to resolve; every emission is byte-identical, so any one of them will do. Secret-shaped keys and inline Bearer/JWT tokens are redacted, the Viya hostname is masked in error/result text, and every field is size-capped.

   Records group by run_id — one per server process — not by MCP session. The protocol is moving to a sessionless model (FastMCP 4 makes it the default) in which session_id is absent or minted per request, so grouping on it would shatter every trace into single-call fragments. Under stdio, one process serves one client, so a run is that client's trace. Under HTTP a run spans every client the process served, and client_name/client_version are the only thing separating them — two users on the same client software share one run_id and one seq counter, which is an accepted limitation of dropping the session key, not something a per-process COLLECTION_LOG_PATH can fix (that splits by process, the axis run_id already covers).

Enabling it

Set the toggle in .env (all options are documented in .env.sample):

COLLECTION_MODE=true
# optional overrides (defaults shown):
# COLLECTION_LOG_PATH=~/.sas-mcp-server/tool-usage.log
# COLLECTION_LOG_RESULTS=failures  # never | failures | always (see below)
# COLLECTION_RUN_TAG=            # free-text label stamped into run_start

Tool results are recorded per COLLECTION_LOG_RESULTS — a tri-state dial: never records only a content-free shape summary (type + key names, e.g. {"_type":"object","_keys":["status","report_id"]}); failures (the default) records full (capped + redacted) result contents only for calls that errored or whose tool declared a failure — the middle ground, since failure diagnostics are the highest-value trace data and rarely carry table rows, and because under never a success and a tool-declared failure are indistinguishable in the log; always records result contents on every call. Arguments, goal, status, error text, and the tool-declared outcome fields are captured in every mode. (true/false still work as aliases for always/never.)

⚠️ Privacy: when enabled, the log captures your tool inputs (e.g. the SAS code and queries you submit) and — in failures/always modes — real result data that may include table rows, SAS listings, and PII. Redaction is heuristic (credential-shaped keys + Bearer/JWT + the Viya hostname) and does not detect PII in data values. Review the log before sharing it. The file is locked to your user (chmod 0600 on POSIX; icacls on Windows, best-effort).

Performance impact

Collection mode is designed to be cheap enough to leave on. Measured on this repo (45 registered tools, FastMCP 3.4.2):

• Prompt tokens. The injected goal field grows the tools/list schema the model sees by roughly +2,400 input tokens (~29%) per turn. Because the tool list is stable within a session it is served from the prompt cache after the first turn (steady-state ≈ +240 tokens/turn), plus ~15–30 output tokens per call for the model to write the goal sentence. This is the only client-visible cost and it applies only while collection mode is enabled.
• Per-call latency. Middleware + logging adds ≈1.4 ms per call at the shape-only default (≈5.3 ms with COLLECTION_LOG_RESULTS=always). The JSONL write is offloaded to a worker thread so it never blocks the event loop. Against real Viya calls (typically hundreds of milliseconds to seconds) this is negligible — the live integration suite passed identically with collection mode off and on, the overhead lost in normal network variance.
• Disk. Roughly 0.5–0.7 KB per tool call at the shape-only default. The log rotates at COLLECTION_MAX_LOG_BYTES (default 10 MiB, ≈16k calls) and keeps COLLECTION_LOG_BACKUPS (default 3) rotated files, so on-disk growth is bounded.

Testing

The project includes two layers of tests: unit tests (fast, no credentials required) and integration tests (run against a real SAS Viya instance).

run_tests.sh vs. running pytest directly — pick by platform. run_tests.sh is a Bash convenience wrapper (it adds the ruff + pyright gates, credential wiring, and JUnit reporting). It runs on Linux/macOS — and on Windows only under Git Bash or WSL. On Windows PowerShell or cmd, use the uv run python -m pytest … commands shown under each mode below. They are cross-platform, do the same test selection, and need no setup beyond uv sync.

Running Unit Tests

Unit tests verify tool schemas, request payloads, and internal logic without making any network calls:

./run_tests.sh                                     # Linux/macOS (also runs ruff + pyright)
uv run python -m pytest -m "not integration" -v    # any platform, incl. Windows PowerShell

This runs the unit suite and deselects the integration tests, which then show up in the summary as e.g. 28 deselected. That is expected — those tests are not meant to run in a unit-only pass. They only execute in the integration modes below, because they need a live Viya instance; there is no flag that "activates" them in a not integration run.

Running Integration Tests

Integration tests call every tool against a live Viya environment. They require credentials, provided via .env or CLI arguments.

uv sync installs everything the integration suite needs, including openpyxl (used to build the Excel upload_data fixture). It lives in the test-formats dependency group, which [tool.uv] default-groups syncs by default — so no extra install step is required.

Full suite (unit + integration) — reads VIYA_ENDPOINT, VIYA_USERNAME, VIYA_PASSWORD from .env:

./run_tests.sh --integration      # Linux/macOS
uv run python -m pytest -v        # any platform

Passing credentials on the command line (wrapper only):

./run_tests.sh --integration \
    --endpoint https://your-viya-server.com \
    --username youruser \
    --password yourpassword

With the direct pytest command, set the same three variables in .env (or export them in your shell) instead.

Integration tests only (skip unit tests):

./run_tests.sh --integration-only                    # Linux/macOS
uv run python -m pytest -m integration --no-cov -v   # any platform

The pytest marker is integration, not integration-only. --integration-only is a flag of the run_tests.sh wrapper; the underlying pytest marker is just integration. Running pytest -m "integration-only" matches no marker and silently deselects all tests (0 selected). Use -m integration.

Why --no-cov? pytest.ini enforces a 90% coverage floor that only the full unit suite reaches. An integration-only run exercises far less code (~65%), so without --no-cov pytest exits non-zero with a coverage failure even though every selected test passed. run_tests.sh --integration-only adds --no-cov for you; add it yourself when calling pytest directly (or use --cov-fail-under=0).

Binary upload formats. The Excel upload_data integration test generates its .xlsx fixture with openpyxl, from the test-formats group that uv sync installs by default (see above). If you deliberately sync without it (e.g. uv sync --no-default-groups), the test importorskips — you'll see it as skipped, not failed. csv, tsv, and file_path/data_format coverage needs no extra deps. Generating a sas7bdat/sashdat fixture requires SAS itself, so those two formats are covered by unit-level payload tests only, not live.

Every one of the 92 tools and 9 prompt templates has an integration test, enforced by the test_every_tool_has_integration_coverage / test_every_prompt_has_integration_coverage guards — adding a new tool or prompt without integration coverage fails the suite. The resource-dependent tests discover real targets on the instance: score_data scores the most recently modified MAS module (discovering a real step and its inputs), and run_ml_project re-runs the most recently modified completed ML project. They skip only if the instance has no such resource at all. Likewise, test_catalog_agents_workflow skips with "No discovery agent named 'Public'" on instances where SAS Information Catalog has no discovery agent named Public configured — an expected skip, not a failure; ask a Viya admin to configure one if you need that test to run.

In CI: the .github/workflows/integration.yml workflow runs this suite on demand (manual dispatch, or by adding the run-integration label to a PR) using repository secrets, and publishes the results back to the PR as a status check, a sticky comment, and a downloadable JUnit artifact. Result files are written to reports/ (git-ignored) and are never committed.

Locally (attach results to a PR yourself): run with --report to write the JUnit XML and a Markdown summary into reports/ (git-ignored), then post them to a PR with the GitHub CLI — no commit, no CI required:

./run_tests.sh --integration-only --report
gh pr comment <PR> --body-file reports/integration-summary.md   # summary table as a comment
gh gist create reports/integration.xml                          # full XML as a linkable gist

GitHub has no API/CLI to attach a binary file to a PR (drag-and-drop upload is browser-only), so the summary is posted as a comment and the raw XML is shared via a gist link or pasted in a collapsed <details> block. To produce the canonical Actions artifact from your machine instead, trigger the workflow remotely: gh workflow run integration.yml.

Test Structure

Contributing

Maintainers are accepting patches and contributions to this project. Please read CONTRIBUTING.md for details about submitting contributions to this project.

License & Attribution

Except for the the contents of the /static folder, this project is licensed under the Apache 2.0 License. Elements in the /static folder are owned by SAS and are not released under an open source license. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration.

Separate commercial licenses for SAS software (e.g., SAS Viya) are not included and are required to use these capabilities with SAS software.

As with any container image, direct and indirect dependencies are governed by their own licenses. Users of the published container image are responsible for ensuring that their use complies with all applicable licenses.

All third-party trademarks referenced belong to their respective owners and are only used here for identification and reference purposes, and not to imply any affiliation or endorsement by the trademark owners.

Third-Party Dependencies

This project requires the following dependencies.