Tools

Brain MCP Server

MCP server providing access to 353K conversation messages with semantic search. For developers who want to query their own conversation history.

Capabilities

• semantic_search() — Find similar messages across 106K embeddings
• unified_search() — Cross-source search (conversations, YouTube, GitHub, markdown)
• thinking_trajectory() — Track concept evolution over time
• query_signature_phrases() — Identify language patterns

Data sources

• 353K raw conversation messages (2023-2025)
• 106K embedded messages with semantic vectors
• 31K YouTube videos (consumption patterns)
• GitHub repos + commits

Setup

Requires: Claude conversation exports, embedding pipeline, SQLite database.

Embeddings Pipeline

Process for converting conversation exports into searchable embeddings. Uses OpenAI's text-embedding-3-small model.

Replication

To build similar infrastructure for your own data:

1. Export conversations (Claude, ChatGPT, etc.)
2. Parse into structured format (JSON)
3. Generate embeddings for semantic search
4. Store in SQLite with FTS5 for hybrid search
5. Expose via MCP for Claude Code integration

Full guide and source code coming.