memory-bridge-mcp 1.2.0

Unified MCP memory server bridging Mind (concept graph) and Honcho (semantic vector) memory with episodic timeline and LLM synthesis

Memory Bridge MCP

Unified MCP memory server — bridges Mind (concept graph) and Honcho (semantic vector) memory with episodic timeline and LLM synthesis into a single MCP query surface.

pip install memory-bridge-mcp

What it does

Memory Bridge exposes 8 MCP tools that let any MCP-compatible agent (Claude Code, Codex, etc.) store and recall memories across three memory systems:

Tool	Purpose
bridge_remember	Store a memory — auto-routes short facts to Mind, long content to Honcho
bridge_recall	Unified recall across both systems with Reciprocal Rank Fusion
bridge_status	Combined health report for all memory systems
bridge_episode_record	Record a structured session episode
bridge_episode_recall	Search episodic timeline by text, time, and tags
bridge_timeline	Chronological view of all session episodes
bridge_synthesize	Cortex Accelerator — synthesize knowledge across all memory
bridge_contradictions	Detect contradictory memories

Quick start

# Install
pip install memory-bridge-mcp

# Run as MCP server (stdio)
memory-bridge

The server speaks the MCP protocol over stdio. Configure it as an MCP server in your agent's config:

{
  "mcpServers": {
    "memory-bridge": {
      "command": "memory-bridge"
    }
  }
}

Memory systems

Mind (concept graph)

A local, zero-dependency concept graph with spreading-activation recall, Ebbinghaus forgetting, and dream consolidation. Stored in .mind/graph.json in the working directory. The vendored mind.py module handles all graph operations.

Honcho (semantic vector)

An optional semantic vector store running on http://127.0.0.1:8000. Provides embedding-based similarity search. If Honcho is unavailable, the bridge falls back to Mind only.

Episodic (session timeline)

A JSONL-based append-only log of structured session episodes. Stored in .memory-bridge/episodes.jsonl. Supports text search, time-range filtering, tag filtering, and session filtering.

Synthesis (Cortex Accelerator)

Optionally uses a local Ollama instance (http://127.0.0.1:11434) to synthesize related memories into concise summaries. Falls back to extractive key-sentence selection when LLM is unavailable.

Configuration

The server reads these environment variables:

Variable	Default	Description
HONCHO_BASE_URL	http://127.0.0.1:8000/v3	Honcho API endpoint
OLLAMA_BASE_URL	http://127.0.0.1:11434	Ollama API endpoint
OLLAMA_SYNTHESIS_MODEL	qwen3:8b	Model for LLM synthesis

Architecture

┌──────────────┐     ┌──────────────┐     ┌──────────────┐
│   Agent      │     │   Agent      │     │   Agent      │
│ (Claude,     │     │ (Codex,      │     │ (any MCP)    │
│  Codex, ...) │     │  ...)        │     │              │
└──────┬───────┘     └──────┬───────┘     └──────┬───────┘
       │                    │                    │
       └────────────────────┼────────────────────┘
                            │ MCP stdio
                    ┌───────▼────────┐
                    │  Memory Bridge  │
                    │  MCP Server     │
                    └───┬────┬────┬──┘
                        │    │    │
              ┌─────────┘    │    └──────────┐
              ▼              ▼                ▼
       ┌──────────┐  ┌──────────┐  ┌──────────────┐
       │   Mind   │  │  Honcho  │  │  Episodic     │
       │  Graph   │  │  Vector  │  │  Timeline     │
       │ (local)  │  │ (HTTP)   │  │  (JSONL)      │
       └──────────┘  └──────────┘  └──────────────┘

Zero dependencies

The core server uses only the Python standard library. No pip dependencies required. The vendored mind.py module is also zero-dependency.

License

MIT