mcp-brain

What it does

mcp-brain acts as a persistent, structured memory layer for coding agents like Claude Code. Instead of forcing the LLM to re-explore the repository every session, it provides a compressed "awareness" of the project, tracking high-signal changes, architectural decisions, and team claims to reduce token waste and improve file localization accuracy.

Tools

• brain_init: Register a new project with its stack and core conventions.
• brain_get_context: Retrieve a compressed L1 YAML context for immediate session orientation.
• brain_predict_files: Given an issue title or body, predict the most relevant files using BM25 and code graph analysis.
• brain_remember: Store a new architectural decision or pattern in the memory hierarchy.
• brain_get_decisions: Recall historical reasoning and trade-offs from L2 memory.
• brain_save_session: Create a structured snapshot of the session's progress.
• brain_start_ticket: Orchestrate ticket workflows with built-in conflict detection.
• brain_record_outcome: Log whether a ticket was completed or reverted to refine memory precision.

Installation

Add the following to your claude_desktop_config.json:

{
  "mcpServers": {
    "mcp-brain": {
      "command": "python",
      "args": ["/absolute/path/to/mcp-brain/run.py"]
    }
  }
}

Supported hosts

• Claude Code
• Claude Desktop