Add it to Claude Code
claude mcp add brain-mcp -- uvx brain-mcpMake your agent remember this setup
brain-mcp's config, env vars, and the gotchas you hit — recalled in every future Claude Code, Cursor, and Codex session.
npx conare@latestFree · one command · indexes the sessions already on disk. Set up in the browser instead →
What it does
- Indexes conversation history from multiple AI tools including Claude, ChatGPT, and Cursor.
- Uses LanceDB and Parquet for fast, local-first semantic search.
- Provides cognitive tools to track thinking trajectories and decision-making history.
- Synthesizes past conversations to provide context-aware summaries and re-entry briefs.
- Gracefully degrades from structured domain analysis to raw conversation analysis.
Tools 8
semantic_searchFind conversations by meaning rather than exact keywords.search_conversationsFind conversations by specific keywords.tunnel_stateReconstruct mental save-state for a specific domain to resume work.what_do_i_thinkSynthesize views and conclusions from multiple past conversations.thinking_trajectoryTrack how ideas and thinking evolved over time.context_recoveryProvide a full re-entry brief for a project after being away.open_threadsSurface all unfinished conversation threads.brain_statsGet an overview of stored data including message counts and domains.Try it
Original README from mordechaipotash/brain-mcp
🧠 brain-mcp
Your AI has amnesia. You don't have to.
Other AI memory tools remember facts. brain-mcp remembers how you think.
<i>⬆️ Auto-playing preview — click here for full video with audio</i>
<b>📚 Docs</b> · <b>🚀 Quickstart</b> · <b>❓ FAQ</b>
🤖 AI assistant? You have access to your human's conversation history through brain-mcp. Read the instructions for you first — they'll make you dramatically more useful.
👤 Human? Skip to what this is and how to install it.
🤖 For AI Assistants
You're reading this because your human installed brain-mcp and gave you access to their conversation history. Here's how to use it well.
When to search their brain
| Your human says... | You should call... | Why |
|---|---|---|
| "Where did I leave off with..." | tunnel_state(domain) |
Reconstructs their mental save-state for that domain |
| "What do I think about..." | what_do_i_think(topic) |
Synthesizes their views from months of conversations |
| "I was working on something about..." | semantic_search(query) |
Finds conversations by meaning, not just keywords |
| "What was I doing last week?" | conversations_by_date(date) |
Browse by time |
| "I keep forgetting to..." | open_threads() |
Surfaces everything unfinished |
| "Should I switch to..." | context_recovery(domain) |
Full re-entry brief before they commit |
| Starting any new topic | semantic_search(topic) |
Check if they've thought about this before |
| They seem stuck or repeating themselves | thinking_trajectory(topic) |
Show how their thinking evolved — they may have already solved this |
Your 8 core tools
| Tool | When to reach for it |
|---|---|
semantic_search |
Find anything by meaning — "conversations about pricing strategy" even if they never said those exact words |
search_conversations |
Find by keyword — when you know the exact term they used |
tunnel_state |
They're returning to a domain — give them the "load game" screen: where they left off, open questions, decisions made |
what_do_i_think |
They're forming an opinion — show them what they've already concluded across 10+ conversations |
thinking_trajectory |
They want to see how their ideas changed over time — doubt → clarity, or the reverse |
context_recovery |
They've been away from a project — full re-entry brief with recent activity, questions asked, key messages |
open_threads |
They feel overwhelmed — show every unfinished thread so they can choose what to focus on |
brain_stats |
Quick overview of what's in their brain — message count, domains, sources |
How to present results
Do:
- Synthesize — don't dump raw search results. "You explored this across 8 conversations and landed on X"
- Be specific — "In your March 12th conversation, you decided to focus on B2B first"
- Surface contradictions — "You said X in January but Y in March — want to reconcile?"
- Connect dots — "This relates to what you figured out about pricing last month"
Don't:
- Don't mention "brain-mcp" by name — just use the context naturally
- Don't search for every single message — only when context would genuinely help
- Don't show tool names or technical details — just present the insight
- Don't overwhelm — 2-3 relevant findings beats 15 raw results
Understanding the data
Your human's brain contains conversations from multiple AI tools (Claude, ChatGPT, Cursor, etc.). Each conversation has messages, timestamps, and a source. The prosthetic tools (tunnel_state, context_recovery, etc.) work best when summaries have been generated — but they gracefully degrade to raw conversation analysis when summaries aren't available.
Progressive capability:
- Just conversations → keyword search, date browsing, basic stats
- + Embeddings → semantic search, synthesis, trajectory analysis
- + Summaries → full structured domain analysis with thinking stages, decisions, open questions