MCP Index Notes

Quick Tip: Copy-Paste Text and Images

You can simply copy and paste text or images into your favorite chat client (like Copilot or Anthropic) and say something like:

Add all this to my notes

The LLM will index the information as it decides, using its own context and capabilities. This makes it easy to capture and organize information without manual formatting or tool calls.

MCP Index Notes

A simple, fast MCP server to index and retrieve notes using SQLite (FTS5) with optional JSON backups. Written in TypeScript with verbose logging.

Features

• Fast local storage using better-sqlite3
• Full-text search (FTS5) across content, key, tags, metadata
• Upsert by id or insert by key
• Query by key or text
• Delete by id or key
• Backup to JSON and restore
• Structured, verbose logging via pino

Tools

• index-upsert: Create/update a note
• index-query: Query by key or full-text search
• index-delete: Delete by id or key
• index-backup: Export all notes to JSON
• index-restore: Import notes from JSON
• index-list-keys: Return keys with counts
• index-health: Health check

Graph tools:

• graph-node-upsert: Create/update a graph node
• graph-neighbors: Get neighbors of a node
• graph-path: Find a path between nodes
• graph-import-from-notes: Build graph from existing notes (note->key, note->tags)
• graph-stats: Node/edge counts

Image tools:

• image-upsert: Store an image (base64 or file path) under a key
• image-get: Retrieve images by id or key (optionally include base64 data)
• image-delete: Delete single image by id or all images under a key
• image-export: Export stored images (by id or key) to files on disk

Quick start

1. Install deps

2. Dev run (stdio MCP server)

3. Smoke test (local DB only)

Environment vars:

• DB_PATH: path for SQLite db (default ./data/notes.db)
• LOG_LEVEL: pino level (trace|debug|info|warn|error)
• LOG_PRETTY: true for human-readable logs

Integrations

Below are ready-to-copy examples for popular MCP hosts and clients. Replace paths with your local ones on Windows. All examples run this server via Node and pass optional env vars.

Tip: Build once so the dist entry exists.

GitHub Copilot Chat (VS Code)

Copilot Chat supports MCP servers. Add a new MCP server entry pointing to your built script.

• Open VS Code Settings (JSON) and add an MCP server entry under the Copilot MCP section (exact setting label may vary by version). Use this structure:

Then in Copilot Chat, ask it to call a tool, e.g.: “Call tool index-health”. You should see { "ok": true } in the result.

Claude Desktop

Add to Claude Desktop’s settings.json (Help → Open config file). Example:

Cursor

Create or edit ~/.cursor/mcp.json:

Continue.dev

Add to ~/.continue/config.json under mcpServers (structure may vary by version):

Verify the connection

Prompt Examples (How to use MCP tools in chat)

Advanced LLM Prompts

You can leverage advanced LLMs to interact with MCP tools for more intelligent workflows. Here are some example prompts:

Semantic Search

Expected result: Returns notes relevant to database security using full-text search.

Summarize Notes

Expected workflow: The client calls index-query with { tags: "meeting" }, then uses the LLM to summarize the returned notes.

Generate Knowledge Graph

Expected workflow: The client calls graph-import-from-notes and graph-stats to visualize relationships between notes and tags.

Find Shortest Path Between Concepts

Expected result: Returns the path of related notes/tags between the two concepts.

Context-Aware Upsert

Expected workflow: The client upserts the note and updates the graph to connect related concepts.

Multi-step Reasoning

Expected workflow: The client queries notes, summarizes with LLM, and generates actionable suggestions.

You can interact with the MCP server using natural language prompts in your chat client. Here are some example prompts:

Health Check

Expected result: { "ok": true }

Add a Note

Expected result: { "id": ... }

Query a Note by Key

Expected result: Returns stored entries for that key

Full-Text Search

Expected result: Returns notes containing the word "connection"

List All Keys

Expected result: List of all note keys with counts

Delete a Note

Expected result: Confirmation of deletion

Backup Notes to JSON

Expected result: JSON export of all notes

Restore Notes from JSON

Expected result: Notes imported from backup

Graph Tools

Expected result: Graph operations as described

Store an Image (base64)

Returns: { "id": ... }

Store an Image (file path)

Get Images By Key (metadata only)

Get Images Including Data

Returns array with base64 data field.

Delete Image

Delete All Images Under a Key

Export Images to Files

Returns list of written file paths.

If your client shows a tools list, you should see all tools from this server.

JSON backup format

Notes

• The MCP server communicates via stdio. Integrate with your LLM runtime that supports MCP.
• The DB uses WAL mode for concurrency and performance.