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.
README
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 start - 7 MCP Tools —
add_decision,log_tech_debt,update_progress,add_context,get_memory,search_memory,reindex_memory - Semantic Search — Embedding-based search via
@huggingface/transformers(all-MiniLM-L6-v2) stored locally in SQLite - Git 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 updates - Session Summary — Inactivity timer writes a session summary to
progress.md - Ollama Fallback — Keyword extraction when Ollama is unavailable
Prerequisites
- Node.js >= 20
- Ollama (optional, for automatic summarization)
ollama pull llama3.2
Installation
In a new project (recommended)
npx @pnientiedt/project-memory-mcp init
This sets up everything automatically:
- Creates
.project-memory/config.yamlwith documented defaults - Adds
project-memoryentry to.mcp.json - Updates
.gitignorewith database/log exclusions - Installs the git post-commit hook
- Pulls
llama3.2if Ollama is running
From source
npm install
npm run build
Register 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 index
Configuration
.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: true
Environment variables override config:
| Variable | Overrides |
|---|---|
PMM_OLLAMA_URL |
ollama.base_url |
PMM_OLLAMA_MODEL |
ollama.model |
PMM_HOOK_PORT |
git.hook_port |
PMM_LOG_LEVEL |
logging.level |
Memory Files
Located in .project-memory/ and versioned in git (except embeddings.db and server.log):
| File | Contents |
|---|---|
decisions.md |
Architectural decisions (ADRs) |
tech_debt.md |
Technical debt |
progress.md |
Project progress & session summaries |
context.md |
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 publish
Project 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 hook
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.