Skip to main content

MCP allowing short text notes to be persisted and retrieved locally for aiding in long context reasoning tasks

Project description

An important pre-read caveat. This project is 100% codex written. I wrote nothing execpt the prompts to execute this. I plan to take a pass at cleaning up the code where needed but as of now the code is rough but working.

Notes MCP

The goal of the notes mcp server is to allow for LLMs to write and read small chunks of information that they think may be important to reference later.

Design

  • The notes will be stored as filestore files with a unique uuid for each file.
  • A summary of the note (or details the model provides on when it would want to use that note again) will be persisted to a local vector store along with a metadata field on the entry containing the uuid to reference back to the file.
  • The vector store is a local Chroma instance backed by an on-device embedding function, persisted under .notes/chroma/.

The system is designed with four core actions to interact with this system

Actions

Write Note

Description: This should create a new note, persist it to disk, and add an entry into the Vector Store.

Args:

  • note: str - the note to persist to disk
  • when_to_use - a summary of the note and when the model would want to use this note
  • tags: list[str] (optional) - metadata tags to aid filtered search

Search Notes

Description: Returns up to five notes matching the query, optionally filtered by tags and score threshold.

Args:

  • query: str - the search query
  • tags: list[str] (optional) - only return notes that share at least one of these tags
  • score_threshold: float (optional, default 0.75) - minimum similarity score required
  • k: int (optional, default 5) - max results to return

Return:

  • results: list of objects each containing note, note_id, score, when_to_use, and tags

Remove Note

Description: Deletes a note from filestore & vectore store

Args:

  • note_id: str - the id of the note to delete

Update Note

Description: Updates an existing note. Will overwrite the entire note with the new content provided, is not diff based

Args:

  • note_id: str - the id of the note to update
  • note: str - the new note content to overwrite the old note

Running the server

TODO: Add uvx note-taker-mcp support.

Requirements: Python 3.10+, uv, and the fastmcp/chromadb dependencies installed via uv sync.

Quick start:

uv sync
uv run notes-mcp

By default, each server process creates its own session-scoped data root under /tmp/notes-<session-id>/, with note bodies in notes/ and the vector index in chroma/. When the server shuts down, this temp directory is deleted. Set NOTES_MCP_DATA_DIR=/custom/path (or pass data_dir to build_server) to use a persistent location instead.

Getting the model to reliably take notes

Reference example.AGENTS.md

Development notes

  • Entry point: notes_mcp/server.py (script name: notes-mcp).
  • Storage layer: notes_mcp/storage.py writes UTF-8 note files keyed by UUID.
  • Vector index: notes_mcp/index.py uses Chroma’s persistent client and default local embedding function.
  • Tests: uv run python -m pytest.

The server runs over stdio via uv run notes-mcp and registers four MCP tools: write_note, search_notes, update_note, and remove_note.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

note_taker_mcp-0.1.2.tar.gz (194.8 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

note_taker_mcp-0.1.2-py3-none-any.whl (10.0 kB view details)

Uploaded Python 3

File details

Details for the file note_taker_mcp-0.1.2.tar.gz.

File metadata

  • Download URL: note_taker_mcp-0.1.2.tar.gz
  • Upload date:
  • Size: 194.8 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.13.5

File hashes

Hashes for note_taker_mcp-0.1.2.tar.gz
Algorithm Hash digest
SHA256 93fe77dd48807009326b36520a3216314a3f75720f607c3d1a442b048a5eef4a
MD5 4f4c404ff40ce559489d80a91a6403ce
BLAKE2b-256 2c730d1aa5d4cbb5c0ae6cb4cd66099784ba3b53aac62fcb90e387adeebf6728

See more details on using hashes here.

File details

Details for the file note_taker_mcp-0.1.2-py3-none-any.whl.

File metadata

  • Download URL: note_taker_mcp-0.1.2-py3-none-any.whl
  • Upload date:
  • Size: 10.0 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.13.5

File hashes

Hashes for note_taker_mcp-0.1.2-py3-none-any.whl
Algorithm Hash digest
SHA256 55bb6f754934245dce86172138153093710c1f4dd3f4b045e8b7a850edb6b963
MD5 504a11fdf8b6b49333d0a7b6bd33500e
BLAKE2b-256 b3137d0673b2bba437da896060281eed127124f83905898d57353491ae3e990d

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page