Model Context Protocol · MCP Server

Give Claude a memory you own.

MemoryIntelligence as an MCP server. Your assistant recalls what matters and saves durable decisions on its own — across Claude Desktop, Claude Code, and Cursor. One command to set up. Your API key stays in your keychain, never in a config file.

Get it on PyPI→ Set up in one command→ Source on GitHub→

What this is

MCP (Model Context Protocol) is the open standard that lets an AI assistant use external tools. This package is an MCP server: wire it in once, and your assistant gains three memory tools — capture, recall, and list — plus a proactive instinct to use them. Memories are stored as structured Unified Memory Objects, not raw transcripts. Not AI magic — structured data, with citations.

Set up in one command

Works on macOS, Linux, and Windows. You'll need an API key — grab one from the developer portal (free during beta).

Install the package

From PyPI. Or run it with no install at all via uvx.

$ pip install memoryintelligence-mcp # or, zero-install: uvx memoryintelligence-mcp

Run one-command setup

Prompts for your API key (hidden input), stores it securely in your OS keychain, wires Claude Desktop & Claude Code for you, and runs a diagnostic.

$ mi-mcp setup

Your key is never written into any config file. The launcher resolves it from your macOS Keychain (or a chmod 600 keyfile) at runtime. Restart your assistant and your memory tools are live.

Just talk

No new habits. Your assistant recalls relevant memories before answering and captures durable decisions on its own. Or ask it directly: "remember that we ship releases on Tuesdays" / "what did I decide about the pricing tiers?"

Prefer to wire it by hand?

mi-mcp setup does this for you. If you'd rather edit the config yourself, add the server below. Note there's no key in the file — the launcher resolves it from your keychain.

Claude Desktop · claude_desktop_config.json

{ "mcpServers": { "memoryintelligence": { "command": "/Users/you/.memoryintelligence/mcp/run-mi-mcp.sh" } } } # point it at the launcher (absolute path) — it resolves your key # from the keychain. `mi-mcp wire` writes this entry for you.

Claude Code · terminal

$ mi-mcp wire # writes the launcher config (recommended) # check it's healthy any time: $ mi-mcp doctor

The tools your assistant gets

Three tools are on by default — all most people need. The full set is opt-in with MI_MCP_FULL=1.

Default surface

Tool	What it does
mi_capture	Save a decision, fact, or preference as a structured memory.
mi_ask	Search your memories in natural language — answers come back with citations.
mi_list	Browse recent memories.

Full surface · MI_MCP_FULL=1

Tool	What it does
mi_explain	See why a memory matched a query.
mi_verify	Verify a memory's provenance chain.
mi_forget	Delete a memory (with a receipt). Flagged destructive.
mi_batch	Capture many items at once.
mi_upload	Capture from a file (PDF text today).
mi_match	Compare two memories for relatedness.
mi_account	Account info and key status.

What works today

This is beta. Here's exactly what's live — we flag what isn't rather than oversell it.

LiveProactive memory — the assistant recalls + captures on its own (Claude Desktop, Claude Code, Cursor)

LiveCapture, recall (with citations), and list — the three core tools

LivePDF text upload (behind MI_MCP_FULL=1)

ComingAudio & image transcription on upload — not yet functional on the backend

PlannedLocal .umo vault — offline-first, owner-encrypted memory files

Resources

PyPI

memoryintelligence-mcp

Install the latest release. Changelog, versions, and files.

GitHub

Source & issues

The server source, README, and bug reports.

Reference

API Reference

The endpoints the MCP tools call. Every parameter, every response.

Portal

Get an API key

Free during beta. No credit card required.

Stop re-explaining yourself to your AI.

Set it up→ Read the docs→