Synapse Give Claude instant knowledge of your codebase — local, private, no cloud.
You open Claude Code and ask: "Add retry logic to my API client."
Without Synapse: Claude asks what your stack is. You paste 200 lines of code. You explain your conventions. You do this again tomorrow.
With Synapse: Claude says: "Your API client in src/api/client.ts uses Axios with your custom RetryConfig interface — matching the pattern you used in payment.ts. Here's the implementation:"
Synapse watches your project folders, indexes them locally, and connects to Claude via MCP. Claude knows your codebase without you saying a word.
Install in 2 minutes pip install synapse-mcp
synapse init C:/Projects/myapp
Then paste one line into your Claude Desktop config. Done.
What it does Watches your project folders — indexes code, markdown, and config files automatically Stores everything locally — no API key required, nothing leaves your machine Connects to Claude via MCP — works with Claude Desktop, Cursor, and any MCP client Updates in real time — re-indexes files as you save them
Supported tools Tool Status Claude Code ✅ Supported (primary target) Cursor ✅ Supported via MCP Continue ✅ Supported via MCP Claude Desktop chat 🔜 Coming in v0.2 Any MCP client ✅ Standard MCP protocol
Supported file types .py .ts .tsx .js .jsx .go .rs .java .kt .rb .php .cs .swift .cpp .c .h .sh .ps1 .sql .tf .md .json .yaml .toml .env .graphql .proto · Dockerfile · Makefile · and more
Connect to Claude Code After running synapse init , register Synapse as an MCP server:
claude mcp add synapse synapse-mcp
That's it. Restart claude and run /mcp to confirm synapse shows connected · 2 tools.
Claude Desktop chat support is planned for v0.2 via SSE transport. Track progress in #12.
CLI reference synapse init # Index a folder (run this first)
synapse status # Show what's indexed
synapse search # Search from the terminal
synapse wipe # Delete the local index
Configuration Synapse works out of the box with no config file required. To customize, create ~/.synapse/config.toml:
[synapse]
watch_paths = ["C:/Projects/myapp", "C:/Projects/clientwork"]
ignore_patterns = ["node_modules", ".git", "dist", "*.pyc"]
[embedding]
provider = "local" # "local" (default, free, private) or "openai"
chunk_size = 512
chunk_overlap = 64
Local embeddings (default): Uses nomic-embed-text-v1.5 via sentence-transformers. Downloads ~400MB on first run. Free. Nothing leaves your machine.
OpenAI embeddings (optional): Set provider = "openai" and OPENAI_API_KEY env var. Slightly better quality for natural language queries. Costs ~$0.00 for personal use.
How it works Your files → Synapse watches → Chunks text → Embeds locally → Stores in LanceDB
↓
Claude Desktop ←── MCP query ←── Synapse MCP server ←── Semantic search
synapse init walks your project and indexes every supported file The file watcher triggers incremental re-indexing on every save When you open Claude Desktop, the synapse-mcp server is available as a connected tool Claude calls recall("your question") automatically before answering Relevant code snippets are injected into Claude's context window
No round trips to external servers. No accounts. No rate limits.
FAQ Does my code leave my machine? No. With the default local embedding provider, every computation happens on your machine. The only network call is the initial model download (~400MB, cached permanently after that).
What if I use the openai provider? Your code text is sent to OpenAI's API for embedding. If you work on sensitive or proprietary code, use the default local provider.
How is this different from Cursor's built-in indexing? Cursor's index lives inside Cursor only. Synapse works with Claude Desktop, Cursor, Continue, and any other MCP-compatible client — all from one index.
How is this different from ChatGPT Memory? ChatGPT Memory is locked to ChatGPT and stored on OpenAI's servers. Synapse works with any AI tool and stores everything locally.
Will it index my .env files or secrets? Yes — it indexes .env files if they're in your watch path. If you don't want this, add .env to the ignore_patterns list in your config.
Can I inspect what's in the index? Yes: synapse status shows counts, and synapse search lets you query it directly.
How do I remove a project from the index? Run synapse wipe to clear everything. Then run synapse init again with only the paths you want.
Requirements Python 3.10+ 1 GB free disk space (for the embedding model cache) Claude Desktop (or any MCP client)
Contributing Synapse is MIT-licensed and welcomes contributions.
Good first issues:
Add support for a new file type in synapse/ingestion/reader.py Improve chunking for a specific language Write integration tests
To contribute:
git clone https://github.com/nrkoka786/synapse
cd synapse
pip install -e ".[dev]"
pytest tests/
See CONTRIBUTING.md for guidelines.
Roadmap Core filesystem indexing Claude Code MCP integration Local embeddings (no API key) File watcher (incremental re-indexing) Claude Desktop chat support (SSE transport) — v0.2 Git log ingestion (commit messages + PR descriptions) Markdown + PDF document indexing synapse search TUI with rich output Windows installer (.exe) macOS App Synapse Cloud (encrypted cross-device sync) — planned Team Synapse (shared org context) — planned
License MIT © 2026 NR · github.com/nrkoka786/synapse