kimi-code-memory-mcp

A Python MCP (Model Context Protocol) bridge that exposes TencentDB Agent Memory as MCP tools for Kimi Code CLI.

What it does

Gives your AI coding assistant long-term memory across sessions:

• L0 - Raw conversation storage

• L1 - Atomic memory facts (auto-extracted)

• L2 - Scene/context blocks (auto-clustered)

• L3 - User persona/profile (auto-generated)

The LLM can recall relevant memories, capture new conversations, and search past interactions.

Related MCP server: UseCortex MCP Server

Architecture

Kimi Code CLI  <---stdio--->  Python MCP Bridge (this repo)
                                   |
                                   | HTTP :8420
                                   v
                          TencentDB Agent Memory Gateway
                          (official npm package, runs locally)

This repo is a thin Python bridge — it forwards 5 MCP tools to the official Gateway via HTTP. The Gateway does all the heavy lifting (L0-L3 extraction, vector search, persona generation).

Quick Start

1. Install Python dependencies

pip install -r requirements.txt

2. Set up the Gateway (one-time)

python setup-gateway.py

This installs the official @tencentdb-agent-memory/memory-tencentdb npm package and tsx into ~/.memory-tencentdb/.

3. Configure credentials

cp .env.example .env
# Edit .env and fill in your API keys

You need:

• LLM API key — any OpenAI-compatible endpoint (SiliconFlow, OpenAI, SenseNova, etc.)

• SiliconFlow API key — for embeddings (BAAI/bge-m3)

4. Start the Gateway

python start-gateway.py

For background/autostart mode:

python start-gateway-background.py

5. Register in Kimi Code

Add to your ~/.kimi-code/mcp.json:

{
  "mcpServers": {
    "tencentdb-memory": {
      "command": "python",
      "args": ["path/to/server.py"]
    }
  }
}

MCP Tools

SKILL.md

Include SKILL.md in your Kimi Code skills directory to teach the LLM when and how to use these memory tools.

Requirements

• Python >= 3.12

• Node.js >= 22.16.0 (for the Gateway)

• An OpenAI-compatible LLM API key

• A SiliconFlow API key (for embeddings)

Acknowledgments

• TencentDB-Agent-Memory by TencentCloud (MIT License)

• FastMCP Python framework

License

MIT — see LICENSE

This project includes modifications based on TencentDB-Agent-Memory by TencentCloud.