imprint-mcp-server

A Model Context Protocol server that exposes Imprint's local data — meeting notes, calendar events, email metadata, daily digests — to Claude Desktop, Claude Code, and any other MCP-compatible client.

Read-only by design. The server never writes to Imprint's database; Imprint remains the source of truth for sync, classification, and writes.

What you can ask Claude

"What did we agree on in the Loyco meeting last week?"

"List my meetings tomorrow with online links."

"How many emails did I get from acme.no in April?"

"Summarize what I worked on this Tuesday."

Claude calls the appropriate tool(s), gets structured JSON back, and answers with full context from your own local Imprint data.

Related MCP server: SQLite MCP Server

Tools

Install

npm install -g imprint-mcp-server

Or use it ad-hoc via npx (recommended — no global install needed):

// ~/Library/Application Support/Claude/claude_desktop_config.json
{
  "mcpServers": {
    "imprint": {
      "command": "npx",
      "args": ["-y", "imprint-mcp-server"]
    }
  }
}

Restart Claude Desktop. You'll see Imprint show up in the tools menu, and Claude will call its tools whenever a question matches.

Custom database location

By default the server looks at the standard Imprint path:

~/Library/Application Support/Imprint/imprint.db

Override via env var if Imprint's DB lives elsewhere:

{
  "mcpServers": {
    "imprint": {
      "command": "npx",
      "args": ["-y", "imprint-mcp-server"],
      "env": { "IMPRINT_DB_PATH": "/path/to/imprint.db" }
    }
  }
}

Privacy

• The server runs locally on your machine. The MCP transport is stdio — there's no network listener.

• The SQLite connection is opened read-only.

• Tool results travel from the MCP server to Claude Desktop via stdio, then Claude Desktop sends them to Anthropic's API for the model to read. So: whatever you tell Claude to look up is sent to the model, just like any other Claude conversation. Use the same judgment you would if you pasted the data into a chat.

• Imprint stores only email metadata (sender domain, subject, timestamp) — no message bodies. The MCP server can't surface what isn't there.

Development

git clone <repo>
cd imprint-mcp-server
npm install
npm run build
npm start

Smoke-test by piping JSON-RPC at it:

(
  echo '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"smoke","version":"1"}}}'
  echo '{"jsonrpc":"2.0","method":"notifications/initialized"}'
  echo '{"jsonrpc":"2.0","id":2,"method":"tools/list"}'
  sleep 0.3
) | node dist/index.js

Releasing

Releases are published to npm via GitHub Actions using OIDC trusted publishing — no long-lived tokens are stored in the repo or secrets.

To cut a release:

# 1. Bump version in package.json
npm version patch  # or minor / major

# 2. Push the tag
git push --follow-tags

The Release to npm workflow runs on tag push, builds, and publishes with --provenance (a signed SLSA attestation linking the tarball to the exact commit + run — visible on the npm package page).

One-time npm setup (already done for imprint-mcp-server)

On npmjs.com → Packages → imprint-mcp-server → Settings → Trusted Publishers, the following GitHub Actions config is registered:

This means npm will accept publishes from .github/workflows/release.yml on the PTrobe/imprint-mcp-server repo without any secret. Tokens with 90-day expirations stop being a maintenance burden.

Roadmap

• Tasks tool (requires Imprint to cache tasks in DB; today they're fetched live per provider)

• Activity-events search (window/app tracking)

• Resources (vs. tools) for browsing notes by folder

• Write tools — e.g. create_meeting_note, complete_task