Connect AI assistants to TestRail: manage test cases, suites, runs, results, and attachments.
Model Context Protocol (MCP) server for the TestRail platform that enables LLMs to manage test artifacts. It connects AI assistants (Claude, Cursor, Windsurf, and others) to TestRail for operating on projects, test cases, and test runs via natural-language conversation.
๐ ๏ธ Key Features
Manage TestRail projects
Search and create test cases
Kick off test runs
Record test results
Attach files
๐ Use Cases
AI-assisted QA workflows that manage TestRail content
Creating and updating test cases through conversational queries
Running test executions and recording outcomes in TestRail
Attaching artifacts to test-related records
โก Developer Benefits
Integration with MCP-based AI assistants
Natural-language control for QA and test automation tasks
Centralized interaction with TestRail resources (projects, cases, runs, results, files)
โ ๏ธ Limitations
Source excerpt does not specify authentication, supported TestRail versions, rate limits, or tool coverage beyond the listed capabilities.
Connect Claude, Cursor, Windsurf, and VS Code to TestRail โ an open-source Model Context Protocol (MCP) server for AI-assisted test management.
Manage TestRail projects, search and create test cases, kick off test runs, record results, and attach files through natural-language conversation with your AI assistant โ or from CI/CD with the bundled testrail-cli. Built for QA engineers and AI-assisted test automation.
The TestRail MCP Server is a free, open-source Model Context Protocol server that gives AI assistants direct, structured access to a TestRail instance through the TestRail API v2. Once configured, an assistant such as Claude Desktop, Cursor, Windsurf, or GitHub Copilot in VS Code can search test cases, draft new ones, start test runs, record results, and upload attachments on your behalf โ without you leaving the chat window.
It exposes 35 tools, runs locally on Node.js 18+ over the MCP stdio transport, and is licensed under Apache 2.0. There is nothing to host or deploy: your MCP client launches it on demand with npx.
No context switching. No tedious copy-pasting. Just ask your AI.
NOTE
Compatibility baseline: tested and validated against TestRail 10.6.2 (API v2). Older TestRail instances (including pre-7.x pagination) are also supported via built-in backward compatibility. TestRail Cloud and self-hosted TestRail Server both work.
Browse projects, test suites, and sections to automatically map your QA organization.
๐ Full Case Management
Fetch, create, update, and bulk-edit test cases with comprehensive custom field support.
โถ๏ธ Actionable Execution
Create test runs, update results by test_id or case_id, attach files, and track statuses.
๐ง Context-Aware AI
Dynamically exposes templates, fields, priorities, and statuses so LLMs generate valid, structured data.
๐ฅ๏ธ CLI & CI/CD Native
Run every tool from bash, GitHub Actions, GitLab CI, or Jenkins with zero LLM overhead.
๐ Least-Privilege Controls
Per-mode permissions plus per-tool allowlisting; destructive deletes are off by default.
๐ Quick Start Guide
1. Obtain your TestRail API key
Navigate to My Settings โ API Keys in TestRail and generate a new key. Copy it immediately โ TestRail shows it only once. The API must also be enabled instance-wide under Administration โ Site Settings โ API.
2. Configure your MCP client
Add the server to your MCP client configuration. The Claude Desktop example is shown below; Cursor, Windsurf, and VS Code use the same pattern (see the collapsible sections).
Add to .vscode/mcp.json. The inputs block keeps your API key out of the file:
json
{"inputs":[{"id":"testrail-api-key","type":"promptString","description":"TestRail API key","password":true}],"servers":{"testrail":{"command":"npx","args":["-y","@uarlouski/testrail-mcp-server@latest"],"env":{"TESTRAIL_INSTANCE_URL":"https://your-instance.testrail.io","TESTRAIL_USERNAME":"your@email.com","TESTRAIL_API_KEY":"${input:testrail-api-key}"}}}}
๐ Other MCP Clients
Any MCP-compliant client can use this server, because it speaks the standard MCP stdio transport. Point your client at the npx command with the required environment variables โ no port, URL, or transport configuration needed.
3. See it in action
Restart your client completely, then turbo-charge your QA workflow by asking your AI assistant:
"List all projects in TestRail to find the latest active project."
"Show me all active users in the project to find the right assignee."
"Show me all test cases in section 5 of project 3."
"Create a comprehensive test case for 'Login Validation' with detailed steps."
"Start a new test run containing cases from section 5."
"Mark test case C1042 as passed with the comment 'verified on staging'."
Full per-client setup instructions, including troubleshooting, are in the Getting Started guide.
๐ฅ๏ธ Command Line Interface (CLI) & CI/CD Automation
In addition to interacting via AI assistants, you can invoke any TestRail tool directly from shell scripts, terminal environments, and automated CI/CD pipelines (GitHub Actions, GitLab CI, Jenkins) using testrail-cli or npx โ with zero LLM overhead.
Deterministic Execution: Returns standard Unix exit codes (0 on success, 1 on error).
Pipeline-Native Output: Emits clean JSON to stdout for piping into tools like jq, while diagnostics and errors go to stderr.
Zero Duplication: Reuses the exact same API client, retry logic, and validation schemas as the MCP server.
Invocation methods
bash
# Method 1: Direct npx subcommand (Recommended)
npx @uarlouski/testrail-mcp-server cli <command> [flags]
# Method 2: Global or local binary
testrail-cli <command> [flags]
# Method 3: Via package runner
npx -p @uarlouski/testrail-mcp-server testrail-cli <command> [flags]
Examples
Query projects (query_project)
bash
# List all active projects
npx @uarlouski/testrail-mcp-server cli query_project --action many
# Query a single project by ID
npx @uarlouski/testrail-mcp-server cli query_project --action one --project_id 1
Report automated test results (add_results_for_cases)
bash
# Submit results by case_id โ what your test framework already knows
npx @uarlouski/testrail-mcp-server cli add_results_for_cases \
--run_id 88 \
--results '[{"case_id":1042,"status_id":1,"comment":"Passed in CI"}]'
Export cases for knowledge base / RAG (export_cases_for_rag)
bash
# Export all cases for a project into Markdown & metadata sidecars
npx @uarlouski/testrail-mcp-server cli export_cases_for_rag \
--project_id 1 \
--output_dir ./rag_exports
# Export specific cases by ID (comma-separated list)
npx @uarlouski/testrail-mcp-server cli export_cases_for_rag \
--case_ids C101,C102,103 \
--output_dir ./rag_exports
Command discovery & flag documentation
bash
# List all available commands
npx @uarlouski/testrail-mcp-server cli --help# Show parameter options for a specific tool
npx @uarlouski/testrail-mcp-server cli query_project --help
See the CLI & CI/CD guide for complete GitHub Actions, GitLab CI, and Jenkins workflows.
โ๏ธ Environment Variables & Security Controls
Variable
Description
Required
Default
TESTRAIL_INSTANCE_URL
Your TestRail instance URL (e.g., https://example.testrail.io)
Set to true to enable Shared Steps management tools
false
TESTRAIL_ENABLE_CASE_HISTORY
Set to true to enable Case History and revision tracking tools
false
TESTRAIL_ENABLE_RAG_TOOLS
Set to true to enable experimental Knowledge Base / RAG export tools (export_cases_for_rag). Subject to breaking API changes.
false
TESTRAIL_ALLOW_WRITE_OPERATIONS
Allow write operations (e.g. adding/updating test cases, test runs, sections)
true
TESTRAIL_ALLOW_READ_OPERATIONS
Allow read operations (e.g. retrieving projects, test cases, templates)
true
TESTRAIL_ALLOW_DELETE_OPERATIONS
Allow delete operations (e.g. deleting cases or shared steps). Enabled strictly via true.
false
TESTRAIL_ENABLE_DEPRECATED_TOOLS
Preserved for backward compatibility with existing host configurations.
true
TESTRAIL_DISABLED_TOOLS
Comma-separated list of specific tool names to disable (e.g., mutate_suite,delete_entity). Fails if invalid tool names are specified.
-
With only the three required credentials, 27 of the 35 tools are registered. Ready-made read-only and least-privilege configurations are in the Configuration guide.
๐งฐ Available Tools
All 35 tools, grouped by area. Each declares a read, write, or delete mode, which the server surfaces to clients as MCP annotations (readOnlyHint, destructiveHint, idempotentHint, openWorldHint).
Which AI assistants and MCP clients are supported?
Any MCP-compliant client, because the server uses the standard MCP stdio transport. Setup is verified with Claude Desktop, Cursor, Windsurf, and VS Code. Other clients follow the same pattern: run npx -y @uarlouski/testrail-mcp-server@latest with three environment variables.
Which TestRail versions are supported?
The server targets TestRail API v2 and is tested against TestRail 10.6.2. Older instances work too โ the client detects whether an endpoint returns a modern paginated response or a legacy bare array, so pre-7.x instances need no configuration. Both TestRail Cloud and self-hosted TestRail Server are supported.
Is it safe to give an AI assistant write access to TestRail?
Access is layered rather than all-or-nothing. Every tool declares a read, write, or delete mode; deletes are disabled unless you explicitly set TESTRAIL_ALLOW_DELETE_OPERATIONS=true. You can disable all writes with TESTRAIL_ALLOW_WRITE_OPERATIONS=false, or block individual tools by name with TESTRAIL_DISABLED_TOOLS. The server runs locally and talks only to your TestRail instance โ there is no telemetry and no third-party service in the middle.
Is it free?
Yes โ Apache 2.0 licensed, with no paid tier, licence key, or usage limit. You need your own TestRail subscription, and whatever your AI assistant costs. The CLI has no LLM cost at all.
Can I use it in CI/CD without an AI assistant?
Yes. The package ships a testrail-cli binary exposing every tool as a subcommand, reusing the same API client, retry logic, and validation schemas. JSON on stdout, diagnostics on stderr, exit code 0 or 1 โ see the CLI guide.
How do I stop the AI from inventing custom fields that don't exist?
It can't. Before add_case or update_case sends anything, the server validates every field key against your instance's real schema (fetched via get_case_fields) and rejects unknown keys. Templates, priorities, statuses, and configurations are all exposed as tools too, so the model looks up correct IDs instead of guessing them.
How do I handle thousands of test cases without blowing the context window?
Pass output_file to get_cases (or output_dir to export_cases_for_rag). The server paginates the full result set, writes raw JSON to disk, and returns only a short summary to the model.