project-memory-mcp
Allows automatic summarization of git changes and session summaries using Ollama models (e.g., llama3.2) for keyword extraction and fallback when unavailable.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@project-memory-mcpAdd decision: use React for frontend"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
project-memory-mcp
Persistent AI knowledge base for Claude via Model Context Protocol — fully offline, no cloud API key required.
Stores architectural decisions, technical debt, project progress, and domain knowledge in versioned Markdown files. Claude reads these files automatically via MCP Resources and writes new entries via MCP Tools.
Features
4 MCP Resources — Claude reads
memory://decisions,memory://tech-debt,memory://progress,memory://contextautomatically at session start7 MCP Tools —
add_decision,log_tech_debt,update_progress,add_context,get_memory,search_memory,reindex_memorySemantic Search — Embedding-based search via
@huggingface/transformers(all-MiniLM-L6-v2) stored locally in SQLiteGit Hook — Post-commit hook sends diff + message to Ollama for automatic summarization and auto-commits the updated memory files
Filesystem Watcher — Changes to
CLAUDE.md,docs/adr/, etc. trigger memory updatesSession Summary — Inactivity timer writes a session summary to
progress.mdOllama Fallback — Keyword extraction when Ollama is unavailable
Related MCP server: mcp-memory-vault
Prerequisites
Node.js >= 20
Ollama (optional, for automatic summarization)
ollama pull llama3.2Installation
In a new project (recommended)
npx @pnientiedt/project-memory-mcp initThis sets up everything automatically:
Creates
.project-memory/config.yamlwith documented defaultsAdds
project-memoryentry to.mcp.jsonUpdates
.gitignorewith database/log exclusionsInstalls the git post-commit hook
Pulls
llama3.2if Ollama is running
From source
npm install
npm run buildRegister the MCP server in Claude Code (.mcp.json already pre-configured):
{
"mcpServers": {
"project-memory": {
"command": "npx",
"args": ["@pnientiedt/project-memory-mcp"]
}
}
}Usage
After startup Claude reads the memory files automatically. New entries are written via tools:
add_decision — Save an architectural decision (ADR format); upsert=true to replace existing
log_tech_debt — Record technical debt; upsert=true to replace existing
update_progress — Update a milestone; upsert=true to replace existing
add_context — Save domain knowledge / conventions; upsert=true to replace existing
get_memory — Read a memory file directly
search_memory — Semantic search across the knowledge base
reindex_memory — Rebuild the embedding indexConfiguration
.project-memory/config.yaml (created with defaults on first run):
ollama:
base_url: "http://localhost:11434"
model: "llama3.2"
timeout_seconds: 60
fallback_to_keywords: true
embeddings:
model: "Xenova/all-MiniLM-L6-v2"
db_path: ".project-memory/embeddings.db"
git:
hook_enabled: true
hook_port: 47832
skip_keyword: "[skip-memory]"
watcher:
enabled: true
paths: ["CLAUDE.md", "docs/adr/", "README.md"]
debounce_ms: 2000
session:
inactivity_timeout_minutes: 30
summarize_on_end: trueEnvironment variables override config:
Variable | Overrides |
|
|
|
|
|
|
|
|
Memory Files
Located in .project-memory/ and versioned in git (except embeddings.db and server.log):
File | Contents |
| Architectural decisions (ADRs) |
| Technical debt |
| Project progress & session summaries |
| Domain knowledge & conventions |
Development
npm test # Run tests (101 tests)
npm run test:coverage # Coverage report (83% line coverage)
npm run build # Compile TypeScript
npm run typecheck # Type check only
npm run release # typecheck + test + build + npm publishProject Structure
src/
index.ts # Entry point, stdio transport
server.ts # MCP server factory
config.ts # Load & validate configuration (zod)
types.ts # Shared TypeScript types
resources/
memory.ts # MCP resources (memory://)
tools/
write.ts # add_decision, log_tech_debt, update_progress, add_context
read.ts # get_memory, search_memory
admin.ts # reindex_memory
services/
file.ts # Atomic read/write of memory files
embedding.ts # transformers.js + cosine similarity
db.ts # SQLite schema (WAL mode)
ollama.ts # Ollama HTTP client + keyword fallback
summarization.ts # Prompts per memory category
git.ts # simple-git integration
triggers/
git-hook.ts # HTTP server for post-commit hook
watcher.ts # Filesystem watcher (chokidar)
session.ts # Inactivity timer & session summary
hooks/
post-commit # Shell script for git hookThis server cannot be deployed
Maintenance
Related MCP Connectors
Persistent context for Claude. Your AI always knows your projects and next actions across sessions.
Persistent, governed institutional memory for Claude Code — specs, decisions, learnings.
Personal context for every AI: search, read, and write back to your private Markdown library.
Give your AI agent persistent, governed memory for every project. At task start it recalls the approved decisions, conventions, risks and architecture (semantic search, ranked by importance); at close it proposes what was learned as typed memories that you review and approve — governance, not a notes dump. Agents propose, humans govern: edits go back to pending and deletion is human-only by design. Connect Claude Code, Cursor, Claude Desktop or any MCP client in two minutes with just your API key — hosted (nothing to install) or locally via `uvx solucortex-mcp`. Built by SoluAI and dogfooded daily: SoluCortex is developed using its own living memory.
Related MCP Servers
- AlicenseNot gradedqualityAmaintenanceA local-first CLI and MCP server that helps you build and search a personal knowledge vault from Markdown notes, with semantic search and AI-powered features like stale note detection and session memory harvesting. It’s provider-agnostic, requires no GPU in its default mode, and exposes your vault as long-term memory to any MCP-compatible AI tool like Claude Code.47Apache 2.0
- AlicenseNot gradedqualityDmaintenanceA self-hosted memory layer for Claude using plain markdown files, enabling persistent context across different Claude accounts.4 npmMIT
- AlicenseNot gradedqualityCmaintenancePersistent knowledge base tools for Claude Code that store project context, decisions, and patterns locally across sessions, eliminating cold starts.MIT
- AlicenseBqualityBmaintenancePersistent memory for AI agents built on the LLM Wiki pattern: a plain-Markdown brain (also a valid Obsidian vault) with SQLite metadata, local semantic search via fastembed (no API keys), one-call session context with project auto-detection, and a decision log with rationale. Works with Claude Code, Claude Desktop, Cursor, and any MCP client.31MIT