# Agent Context Bridge

> Agent Context Bridge is an open-source MCP server for agent tooling. Shares coding-agent project decisions across Claude Code, Hermes, OpenCode and Codex. It runs locally on your machine and is started by your AI client.

- **Kind:** MCP server
- **Pricing:** Open source
- **Platforms:** CLI, macOS, Linux, Windows, Self-hosted
- **Repository:** https://github.com/shenn667/agent-context-bridge
- **MCP Registry name:** io.github.shenn667/agent-context-bridge
- **Categories:** [agent tooling](https://ai-powered.com/categories/agent-tooling)
- **Listing updated:** 2026-11-19

## About

The README says the bridge records project decisions, architectural constraints and task progress once, then replays them to each coding agent in that agent's own format. Storage is a local folder inside the project (.agent_bridge/) holding a JSONL log and an auto-maintained markdown summary, with no daemon, database or network calls.

For Claude Code and Claude Desktop, the separate acb-mcp stdio server exposes two read-only tools, acb_search and acb_status. Pruning and integrity audits are CLI-only and never run through MCP. Hooks can be installed for Claude Code so context is injected at session start.

It suits developers who switch between several coding agents on the same repository and want earlier decisions to carry over without agents re-reading the whole codebase. The README reports a 98.7% token reduction in one internal benchmark on a 30-record store.

## Features

- **Shared context store:** Records decisions and task notes in a file-based store inside the project folder.
- **Per-agent injection:** Formats stored context for Hermes, Claude Code, OpenCode and Codex adapters.
- **Read-only MCP tools:** The acb-mcp server offers acb_search and acb_status, both read-only, for Claude clients.
- **Offline, zero dependencies:** Uses only the Python standard library and runs without a network connection.
- **Store health commands:** Provides status, integrity audit and keep-last pruning commands from the CLI.

## Good for

- Pass architectural decisions from one coding agent to the next within a repository
- Search earlier agent notes from Claude Desktop or Claude Code over MCP
- Check the health and size of a shared context store before trimming old records
- Reduce repeated repository re-reads when several coding agents work on the same project

## Install

### Claude Code (Terminal)

```sh
claude mcp add agent-context-bridge -- uvx agent-context-bridge
```

### Claude Desktop (claude_desktop_config.json)

```json
{
  "mcpServers": {
    "agent-context-bridge": {
      "command": "uvx",
      "args": [
        "agent-context-bridge"
      ]
    }
  }
}
```

### Cursor (~/.cursor/mcp.json)

```json
{
  "mcpServers": {
    "agent-context-bridge": {
      "command": "uvx",
      "args": [
        "agent-context-bridge"
      ]
    }
  }
}
```

### VS Code (.vscode/mcp.json)

```json
{
  "servers": {
    "agent-context-bridge": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "agent-context-bridge"
      ]
    }
  }
}
```

### Windsurf (~/.codeium/windsurf/mcp_config.json)

```json
{
  "mcpServers": {
    "agent-context-bridge": {
      "command": "uvx",
      "args": [
        "agent-context-bridge"
      ]
    }
  }
}
```

### Other clients (your client's MCP settings (most use this shape))

```json
{
  "mcpServers": {
    "agent-context-bridge": {
      "command": "uvx",
      "args": [
        "agent-context-bridge"
      ]
    }
  }
}
```

## Tools

- `acb_search`: Searches shared context records (read-only).
- `acb_status`: Reports store health (read-only).

## FAQ

### Does the MCP server modify the store?

The README says the two MCP tools, acb_search and acb_status, are strictly read-only, and that prune runs only from the CLI.

### Does Agent Context Bridge need an internet connection or a database?

No. The README states it is offline and local-first, with no daemon, no database and no network.

### How is Agent Context Bridge installed?

The README gives pip install agent-context-bridge, or a pip install -e from a cloned repository.

### Is Agent Context Bridge free?

Yes. Agent Context Bridge is open source, so you can use and self-host it for free.

### How do I add Agent Context Bridge to Claude Code?

Run this command in your terminal: claude mcp add agent-context-bridge -- uvx agent-context-bridge

### What are the best alternatives to Agent Context Bridge?

The closest alternatives to Agent Context Bridge are Agent Memory MCP, databasedmcp, Seamless, Agentrica and DeskPartner.

## Where this comes from

- Description: written from the project's README and its MCP Registry entry.
- Install steps: from the official MCP Registry entry.

## Alternatives

- [Agent Memory MCP](https://ai-powered.com/mcp/agent-memory-mcp): Agent Memory MCP is an open-source MCP server for knowledge and memory.
- [databasedmcp](https://ai-powered.com/mcp/databasedmcp): databasedmcp is a freemium MCP server for databases.
- [Seamless](https://ai-powered.com/mcp/seamless): Seamless is an open-source MCP server for agent tooling.
- [Agentrica](https://ai-powered.com/mcp/agentrica): Agentrica is an open-source MCP server for agent tooling.
- [DeskPartner](https://ai-powered.com/mcp/deskpartner): DeskPartner is an open-source MCP server for agent tooling.
- [agent-exporter MCP bridge](https://ai-powered.com/mcp/agent-exporter-mcp-bridge): agent-exporter MCP bridge is an open-source MCP server for developer tools.
- [Senrelay (ContextCore)](https://ai-powered.com/mcp/senrelay-contextcore): Senrelay (ContextCore) is an open-source MCP server for agent tooling.
- [OpenCrawl](https://ai-powered.com/mcp/opencrawl): OpenCrawl is a freemium MCP server for agent tooling.

---
Source: https://ai-powered.com/mcp/agent-context-bridge
