Inkwell — Persistent Knowledge Base MCP Server

Inkwell is a self-hosted Model Context Protocol server that gives AI agents persistent memory across sessions and projects. Three things set it apart from the other memory servers: plain Markdown files are the source of truth (the index is a throwaway cache you can delete and rebuild), entries are connected by a typed kb:// graph rather than dumped into a flat pile, and updates are bi-temporal — superseding a fact keeps the old version readable instead of overwriting it.

Docker Hub: foreigndmitryi/inkwell-memory · Website: veronchenko.github.io/inkwell-memory

Formerly Engram. Renamed in 0.14.0 — there are a dozen unrelated projects called "Engram" and the name had stopped being findable. Not affiliated with any of them, nor with the Engram keyboard layout. See the changelog for what the rename breaks.

Table of Contents

• Concept
• Features
• Comparison
• Measured Against a Plain Markdown Wiki
• Design Patterns
• Architecture
• Quick Start
• Search
• Tools
• Dashboard
• Graph Relations
• Usage Examples
• Prompt Your Agent
• Configuration
• Storage Format
• Development
• License

Concept

Agent conversations end and take their context with them. Inkwell is the piece that survives: a knowledge base an agent searches before acting and writes to after resolving something non-obvious, so the next session — same project or a different one — starts with what was already learned instead of re-deriving it.

It deliberately stores zero discoverable information. If a fact can be pulled from code, git history, config files, or existing docs, it does not belong in Inkwell — that's what greps and re-reads are for. What belongs is the kind of knowledge a conversation would otherwise lose: a decision and the alternatives it ruled out, a bug's root cause and fix, a procedure learned the hard way, a preference stated once that should hold from then on.

Two things keep the base usable as it grows:

• Atomicity — one entry, one fact. remember warns (non-blocking) on Markdown headers, more than 3 paragraphs, or content past 512 B/1 KB, pushing multi-fact dumps back into separate linked entries instead of a wall of text no search will rank well.
• The graph, not a pile — entries link to each other via kb://uuid#type references, so related facts (a hub project, its features, a diagnostic tied to one of them) stay navigable both ways instead of living as isolated rows.

Features

• Hybrid search — SQLite FTS5 (BM25, Porter stemming) fused with cosine similarity over local Model2Vec embeddings and an IDF-weighted exact-match channel over title/tags via Reciprocal Rank Fusion; finds entries by meaning or by a literal proper noun BM25/embeddings alone would dilute among lexically-similar distractors, with zero cloud dependency
• Typed graph relations — kb://uuid#type links between entries, resolved both directions (outgoing + backlinks) on every recall, with an optional second hop (hops=2) to see how two entries connect through an intermediate one
• Schema-enforced entry types — hub, decision, diagnostic, feature, procedure, integration, pattern, snippet, preference, idea — declared in schema.json and exposed to the client as an enum, so an invalid type can't be written; filterable on search/list
• doctor integrity pass — one schema-driven check over the Markdown files for dangling and superseded links, undeclared types, missing template fields, supernodes, and tag/type collisions
• part_of structural membership — links a detail entry (decision, diagnostic, feature, procedure, integration, ...) to its hub, enforced per type by the schema; filterable on search/list, grouped alongside kb:// back-links in a hub's recall digest
• Bi-temporal versioning — remember(..., supersede=True) creates a new version instead of overwriting; old versions stay in history (include_superseded=True) instead of being lost
• Duplicate detection & link suggestions — remember matches near-identical titles to avoid duplicate entries, and returns suggested_links (embedding-similarity matches) so related facts get cross-referenced instead of orphaned
• Atomicity guardrails — non-blocking warnings on structural anti-patterns (headers, >3 paragraphs, oversized content) so the base stays one-fact-per-entry as it scales
• Web dashboard — force-directed graph view, hybrid search, and a CRUD panel over the same knowledge base the MCP tools use (see Dashboard)
• Three transports — stdio (agent-managed), SSE, streamable-http — so the same server works for a single local agent or a shared multi-agent deployment
• Markdown as source of truth — the SQLite index is a rebuildable cache; delete it and rebuild, no data is ever lost

Comparison

Inkwell trades automatic extraction (Mem0, Zep) for an agent-curated, atomic, explicitly-linked knowledge base — no LLM-driven ingestion pipeline, no graph database dependency, and the on-disk Markdown stays human-readable and diffable.

Measured Against a Plain Markdown Wiki

Inkwell was benchmarked against the same knowledge base packaged as an ordinary Markdown wiki — one file per topic, organized in folders by project and category (decision, diagnostic, feature, procedure, ...), each project with an index page and cross-links between related pages, browsed with Read/Grep/Glob/Bash. Same content, two ways of finding it — for equivalent fact coverage, Inkwell's search/recall used:

32 questions (single-hop, multi-hop, negative, cross-lingual, supersede) over 5,070 entries — 70 curated facts across 4 fictional projects plus 5,000 real-text distractor entries, so retrieval has to work at a scale that doesn't fit in an agent's context:

wiki/
├── Ledgerbird/                       )
├── Pipewren/                         )  4 curated projects — 70 real
├── Snipfox/                          )  decisions/diagnostics/features/
├── Featherstore/                     )  procedures/integrations/snippets
│   ├── README.md                        <- project index page, links to every entry below
│   ├── decision/
│   │   ├── offline-engine-duckdb-over-spark.md
│   │   ├── online-value-serialization-msgpack.md
│   │   └── ... (2 more)
│   ├── diagnostic/
│   │   ├── redis-memory-doubling-from-ttl-less-deprecated-feature-groups.md
│   │   └── training-serving-skew-from-tz-naive-event-timestamps.md
│   ├── feature/       (4 entries)
│   ├── integration/   (2 entries)
│   ├── procedure/     (1 entry)
│   └── snippet/       (1 entry)
├── _shared/                             cross-project patterns & preferences
└── haystack-project-00000.../00199/     200 distractor projects x 25 pages
    │                                     = 5,000 real-text (Wikipedia) pages
    ├── README.md                        <- same index-page shape as a real project
    ├── decision/   (2 entries)
    ├── feature/    (3 entries)
    ├── idea/       (7 entries)
    ├── pattern/    (4 entries)
    ├── procedure/  (1 entry)
    └── snippet/    (2 entries)

Every category folder is a flat list of one-file-per-entry, and every project folder (real or distractor) has its own README.md index page linking to all of them — structurally identical, so the wiki arm can't tell curated fact from distractor by shape alone.

An agent that already knows where to look doesn't need to grep, re-read, and re-grep its way there — the efficiency edge holds for equivalent fact coverage. Retrieval ranking has also been improving: adding an IDF-weighted exact-match channel (below) moved MRR from 0.851 to 0.857 and recall@5 from 0.851 to 0.869, with no regression across any language.

Design Patterns

• Reciprocal Rank Fusion — BM25 and embedding rankings are computed independently and merged by rank position rather than raw score, avoiding the need to normalize incomparable similarity metrics.
• Rebuildable cache over source of truth — the SQLite index is fully derived from the Markdown files (rebuild regenerates it from scratch); the database is never the only copy of a fact.
• Bi-temporal versioning — supersede writes a new entry and points the old one at it via superseded_by, instead of overwriting in place, so history stays queryable (include_superseded).
• HATEOAS-style graph traversal — every recall response carries its own outgoing/incoming kb:// links, so navigating the knowledge graph doesn't require a separate query per hop.
• Shared domain layer, two transports — the dashboard's REST API and the MCP tools both call the same KnowledgeBase/SQLiteBackend methods, so there is exactly one code path for writes regardless of which surface triggered them.

Architecture

Markdown files  --->  Search index  --->  MCP  --->  Agent
(source of truth)     (SQLite FTS5 +      (server.py)  (Claude Code,
                        Model2Vec,                       ChatGPT, ...)
                        rebuildable cache)

Markdown files in <data-path>/entries/ are the only source of truth. The SQLite index (search_backend.py) is a disposable cache built from them — BM25 + Model2Vec embeddings fused via Reciprocal Rank Fusion, plus the kb:// relation graph — and can always be regenerated with rebuild. server.py exposes that index to an agent as MCP tools (remember/recall/search/...); the dashboard is an alternate entry point at the same layer, hitting the same KnowledgeBase/index directly over REST instead of MCP, so remember/delete behave identically whether called by an agent or edited by hand in the browser.

• src/server.py — MCP tool definitions (remember, recall, search, list, tags, forget, rebuild, doctor), one process, stdio/SSE/streamable-http transport
• src/database.py — KnowledgeBase: Markdown + YAML frontmatter CRUD, UUID assignment, duplicate detection, bi-temporal supersede logic
• src/schema.json + src/schema.py — the entry taxonomy and its per-type rules as data, plus the loader that turns them into the entry_type enum the MCP client validates against
• src/doctor.py — the schema-driven integrity pass shared by the doctor tool, rebuild's warnings, and remember's conformance check
• src/search_backend.py — SQLiteBackend: BM25 (FTS5) fused with Model2Vec cosine similarity via Reciprocal Rank Fusion, plus kb:// relation extraction/graph traversal; embeddings are computed lazily on write and stored as a BLOB column
• src/dashboard/ — a second, optional process (app.py FastAPI REST + /api/graph, static/index.html vanilla-JS canvas graph, __main__.py its own uvicorn entry point) reusing the same KnowledgeBase
• plugins/inkwell-hooks/ — a self-contained plugin installable in both Claude Code and Codex (.claude-plugin/plugin.json, SessionStart/Stop/SessionEnd + a PreToolUse search-before-remember gate, plus a Claude-Code-only inkwell-project-onboarder agent) that mechanically enforces the search-first, remember-after workflow instead of relying on a system prompt alone; listed as inkwell-hooks in both the repo-root .claude-plugin/marketplace.json (Claude Code) and plugins/marketplace.json (Codex)

In Docker, the MCP server and the dashboard run as two processes in one container (docker-entrypoint.sh), sharing the same /knowledge volume; the container exits if either process dies.

Quick Start

stdio

Your agent manages the server. Recommended for Claude Code, ChatGPT Desktop, Cursor.

claude mcp add --transport stdio inkwell -- \
  docker run -i --rm -v ./knowledge:/knowledge foreigndmitryi/inkwell-memory

SSE

Persistent server on the network. Share knowledge across multiple agents.

docker run -d --name inkwell \
  -p 8192 \
  -v ./knowledge:/knowledge \
  foreigndmitryi/inkwell-memory --transport sse

docker port inkwell 8192   # host port Docker assigned
claude mcp add --transport sse inkwell http://your-host:<port>/sse

-p 8192 (host port omitted) has Docker pick a free ephemeral host port instead of failing when a fixed port like 8192 is already taken by another container — check the actual assignment with docker port. Use -p 8192:8192 instead if you need the host port to stay fixed.

HTTP

Stateless, load-balanceable.

docker run -d --name inkwell \
  -p 8192 \
  -v ./knowledge:/knowledge \
  foreigndmitryi/inkwell-memory --transport streamable-http

docker port inkwell 8192   # host port Docker assigned
claude mcp add --transport http inkwell http://your-host:<port>/mcp

Multi-tenant

One server, several teams, each isolated to its own data folder and API key. Set INKWELL_MULTI_TENANT=1 (requires INKWELL_PUBLIC_URL and INKWELL_ADMIN_API_KEY, and a network transport — never stdio); src/app.py then serves MCP, the admin UI, and the dashboard from one process. Provision a team with inkwell add-team <name> (run via docker exec, see src/cli.py) to get back its API key, then point each team's agent at /mcp with that key as a bearer token:

claude mcp add --transport http inkwell https://your-host/mcp \
  --header "Authorization: Bearer <TEAM_API_KEY>"

Or as a raw mcpServers config (Claude Desktop and other clients):

{
  "mcpServers": {
    "inkwell": {
      "type": "http",
      "url": "https://your-host/mcp",
      "headers": {
        "Authorization": "Bearer <TEAM_API_KEY>"
      }
    }
  }
}

TeamTokenVerifier (src/server.py) hashes the token and looks it up in admin.db on every request — no caching, so a revoked key stops working immediately.

Search

Hybrid: SQLite FTS5 (Porter stemming, BM25) fused with cosine similarity over local Model2Vec embeddings (minishlab/potion-multilingual-128M, no cloud dependency) via Reciprocal Rank Fusion — so a query that shares no literal words with an entry can still find it by meaning. Falls back to keyword-only if the embedding model can't load. The index is still a single SQLite file you can query with standard SQL tools.

Tools

Entry schema

The entry taxonomy and its per-type rules live in schema.json, not in Python: which frontmatter fields a type requires, which template body fields it should carry, whether search may boost it by usage, and whether recall digests its back-links. entry_type is exposed to the MCP client as an enum built from that schema, so an undeclared type is rejected before the call reaches the server.

Resolution order, first found wins: <data-path>/schema.json, then the packaged src/schema.json. A user file replaces the packaged one outright — the two are never merged, so copy the default and edit it. The schema is read once at startup, so editing it requires restarting the server. Entries written by hand can still carry an undeclared type; doctor reports those.

Dashboard

A web UI for visually exploring and hand-editing the same knowledge base the MCP tools use — no protocol duplication, every action goes through KnowledgeBase.

• Force-directed graph of all entries and their kb:// relations — click a node to inspect it, click a legend chip to filter by type (matches stay lit, others dim)
• Hybrid search over the same BM25 + semantic index as the search MCP tool, with tag and entry_type filters
• CRUD panel to create, edit, supersede, or delete entries without touching Markdown files by hand — outgoing/incoming graph relations shown alongside the fields
• Dark, dense, no-chrome interface — a maintenance tool, not a consumer app (see PRODUCT.md/DESIGN.md)

Disabled by default (the container only runs the MCP server). Enable it as a second process in the same container:

docker run -d --name inkwell \
  -e INKWELL_ENABLE_DASHBOARD=1 \
  -p 8192 -p 8193 \
  -v ./knowledge:/knowledge \
  foreigndmitryi/inkwell-memory --transport sse

docker port inkwell 8193   # dashboard host port

Or run it standalone locally: python -m dashboard (see Configuration for INKWELL_DASHBOARD_HOST/INKWELL_DASHBOARD_PORT).

Graph Relations

Link entries with kb://uuid#type URLs in Markdown content:

This service runs on [Saturn](kb://a1b2c3d4-...#runs-on)
and depends on [PostgreSQL](kb://f9e8d7c6-...#depends-on).

recall returns both directions, plus size metadata:

{
  "id": "a1b2c3d4-...",
  "title": "My API Service",
  "content": "...",
  "tags": ["..."],
  "size": 1024,
  "last_modified": "2026-03-14",
  "relations": {
    "out": [{"type": "runs-on", "id": "e5f6...", "title": "Saturn"}],
    "in": [{"type": "depends-on", "id": "b7c8...", "title": "Frontend App"}]
  }
}

Like HATEOAS for knowledge — every response carries the links to navigate the graph.

Usage Examples

Store knowledge

Ask your agent:

"Remember that our API runs on port 8080 and depends on PostgreSQL 15."

Inkwell creates a Markdown file with a unique UUID, indexes it, and confirms. The agent can now recall this fact in any future session.

Search

"What do we know about PostgreSQL?"

Inkwell searches across all entries by content, title, and tags. Results are ranked by relevance.

Navigate the graph

"What depends on PostgreSQL?"

If entries link to the PostgreSQL article with kb://uuid#depends-on, Inkwell returns all backlinks — showing every service that depends on it, without the agent having to search for each one.

Share knowledge across agents

Start Inkwell with SSE or HTTP transport. Multiple agents — even from different providers (Claude, ChatGPT, Copilot) — connect to the same server. What one agent remembers, all others can recall.

Agent A: "Remember that the deploy key rotates every 90 days."
Agent B: "When does the deploy key expire?"
→ Agent B finds the answer immediately.

Prompt Your Agent

Add this to your system prompt or project instructions to make your agent use Inkwell as a reflex, not an afterthought:

Inkwell is your persistent memory. Using it is mandatory, not optional.

Inkwell stores ZERO discoverable information. If you can derive it from
code, git history, configuration files, or existing documentation, it
does not belong in Inkwell. Inkwell captures decisions and their context,
diagnostics and their root causes, procedures learned the hard way —
the kind of knowledge that is lost when a conversation ends.

Before working on any topic: search Inkwell first. Always. Even if you think you know.
Before answering a question about infrastructure or architecture: search first.
Before proposing a solution: check if a past decision exists in Inkwell.

After resolving a diagnostic: remember the root cause and the fix.
After executing a procedure: remember the steps.
After making an architecture decision: remember the choice and the rationale.
After discovering something about the infrastructure: remember it.

A system prompt is easy to forget mid-session. plugins/inkwell-hooks/ ships a self-contained plugin (SessionStart, Stop, SessionEnd, PreToolUse handlers, plus a Claude-Code-only inkwell-project-onboarder agent) that mechanically nudges the agent to search Inkwell before starting work and reminds it to remember non-trivial changes before finishing, instead of relying on it recalling this section unprompted. Works in both Claude Code and Codex — install with /plugin marketplace add <this-repo> then /plugin install inkwell-hooks@inkwell-memory (Claude Code), or codex plugin marketplace add <this-repo>/plugins then codex plugin add inkwell-hooks@inkwell-memory (Codex) — see plugins/inkwell-hooks/README.md for what each hook does and the full install flow.

Disable Claude Code's built-in auto memory

Claude Code ships its own automatic memory (MEMORY.md under ~/.claude/projects/<project>/memory/, loaded every session). Running it alongside Inkwell means two systems writing overlapping notes and competing for the agent's attention, which gets in the way more than it helps. Turn it off in settings.json:

{
  "autoMemoryEnabled": false
}

Or via environment variable (takes precedence over the setting and the /memory toggle): CLAUDE_CODE_DISABLE_AUTO_MEMORY=1. See Claude Code's memory docs for details.

Configuration

All options have INKWELL_* environment variable fallbacks. CLI args take priority.

Storage Format

Entries are Markdown files with YAML frontmatter in <data-path>/entries/:

---
id: a1b2c3d4-e5f6-7890-abcd-ef1234567890
title: Entry Title
tags: [infrastructure, postgresql]
type: decision
resource: /path/to/relevant/file
---

Markdown content here...

type is required on every remember call (a dedicated field, not part of tags) — it classifies the entry (hub, decision, diagnostic, procedure, preference, snippet, ...) and is filterable via search/list. resource is optional: a canonical file/folder path the entry describes. Legacy entries written before these fields existed still read fine.

The search index is a rebuildable cache at <data-path>/index/inkwell.db. Delete it and rebuild — no data is ever lost. rebuild also reports schema-conformance warnings (missing type, malformed resource) across existing entries.

Development

# Build
docker build -t inkwell .

# Test (separate Dockerfile — pytest/tests/ never ship in the production image)
docker build -f tests/Dockerfile -t inkwell-test .
docker run --rm inkwell-test

# Run locally (SSE)
docker run -d --name inkwell -p 8192 -v ./knowledge:/knowledge inkwell --transport sse

License

MIT