https://github.com/catesandrew/mcp-agent-bridge
Bridge AI coding agents (Claude Code, Codex, Copilot) as MCP servers for cross-agent code review and collaboration
https://github.com/catesandrew/mcp-agent-bridge
ai-agents claude code-review codex copilot mcp model-context-protocol typescript
Last synced: about 6 hours ago
JSON representation
Bridge AI coding agents (Claude Code, Codex, Copilot) as MCP servers for cross-agent code review and collaboration
- Host: GitHub
- URL: https://github.com/catesandrew/mcp-agent-bridge
- Owner: catesandrew
- License: apache-2.0
- Created: 2026-05-06T14:31:37.000Z (about 2 months ago)
- Default Branch: main
- Last Pushed: 2026-05-16T18:27:54.000Z (about 1 month ago)
- Last Synced: 2026-05-16T20:36:16.658Z (about 1 month ago)
- Topics: ai-agents, claude, code-review, codex, copilot, mcp, model-context-protocol, typescript
- Language: TypeScript
- Homepage: https://catesandrew.github.io/mcp-agent-bridge/
- Size: 312 KB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- Contributing: CONTRIBUTING.md
- License: LICENSE
Awesome Lists containing this project
README
# MCP Agent Bridge
[](https://github.com/catesandrew/mcp-agent-bridge/actions/workflows/ci.yml)
[](LICENSE)
[](https://nodejs.org)
Bridge AI coding agents (Claude Code, OpenAI Codex, GitHub Copilot) as [MCP](https://modelcontextprotocol.io) servers so they can call each other for code reviews, analysis, and cross-validation.
**[Documentation](https://catesandrew.github.io/mcp-agent-bridge/)**
## Why?
AI coding agents are powerful individually, but they each have blind spots. MCP Agent Bridge lets you:
- **Get a second opinion** -- Send your diff to a different AI model for review
- **Cross-validate** -- Compare responses from Claude, Codex, and Copilot on the same code
- **Structured reviews** -- Get typed JSON output with verdict, issues by severity, and suggestions
- **Share instances** -- Run each agent once as a background service, shared by all clients
## Quick Start
```bash
# Clone and build
git clone https://github.com/catesandrew/mcp-agent-bridge.git
cd mcp-agent-bridge
bun install && bun run build:exe
# Start the Claude bridge on port 8940
npx mcp-proxy --port 8940 -- ./exe/claude-mcp-server
```
Configure your MCP client (`.mcp.json`):
```json
{
"mcpServers": {
"claude_reviewer": {
"type": "streamable-http",
"url": "http://localhost:8940/mcp"
}
}
}
```
Then ask your agent: *"Use the claude_reviewer MCP to review this diff."*
## Architecture
Each bridge runs as a singleton background service behind an HTTP proxy. Multiple clients share one instance.
```
┌─────────────┐ HTTP/SSE ┌───────────┐ stdio ┌──────────────────┐
│ MCP Client │ ───────────────► │ mcp-proxy │ ────────────► │ claude -p │
│ (any agent)│ :8940/mcp │ │ │ codex exec │
└─────────────┘ └───────────┘ │ copilot -p │
└──────────────────┘
```
| Server | Port | Core Tools | Resume Tools |
|--------|------|------------|--------------|
| Claude | 8940 | `review`, `ask`, `code_review` | 16 career tools |
| Codex | 8941 | `codex`, `code_review`, `codex_reply` | 16 career tools |
| Copilot | 8945 | `ask`, `code_review` | 16 career tools |
## Resume & Career Tools
All three servers include 16 resume and career tools powered by embedded skill methodologies. No network fetch at runtime — the expert logic is compiled in.
| Tool | Purpose |
|------|---------|
| `career_fact_extractor` | Extract a fact-ID-tagged career database from any source material |
| `cover_letter_generator` | Personalized cover letters with match analysis and talking points |
| `creative_portfolio_resume` | ATS + designed resume for creative professionals |
| `executive_resume_writer` | C-suite, VP, and Director-level transformation-story resumes |
| `interview_prep_generator` | STAR-method prep with predicted questions ranked by probability |
| `job_description_analyzer` | Match scoring, keyword extraction, and red flag detection |
| `linkedin_profile_optimizer` | Algorithm-aware LinkedIn profile rewrite |
| `portfolio_case_study_writer` | Six-section case studies for design, engineering, PM, and marketing |
| `reference_list_builder` | Reference strategy, formatting, and briefing email templates |
| `resume_ats_optimizer` | Keyword matching and ATS compliance fixes |
| `resume_bullet_writer` | Transform duty bullets into achievement bullets using X-Y-Z formula |
| `resume_formatter` | ATS-safe formatting audit and repair |
| `resume_quantifier` | Add metrics to every bullet using estimation methodology |
| `resume_section_builder` | Build targeted sections by career stage |
| `resume_tailor` | Tailor a master resume to a specific job posting |
| `recruiter_first_screen_simulation` | Simulate a skeptical hiring manager's 45-second resume screen |
| `resume_version_manager` | Organize and track multiple resume versions with naming conventions |
| `tech_resume_optimizer` | Optimize for software engineering, data, DevOps, and technical PM roles |
See the [Resume Tools Guide](https://catesandrew.github.io/mcp-agent-bridge/docs/guides/resume-tools) and [SKILLS.md](SKILLS.md) for full input schemas and usage examples.
## Structured Review Output
The `review` and `code_review` tools return predictable JSON:
```json
{
"verdict": "NEEDS_REVISION",
"issues": [
{
"severity": "critical",
"description": "SQL injection via string interpolation",
"recommendation": "Use parameterized queries"
}
],
"suggestions": [
"Add rate limiting to the login endpoint"
]
}
```
## Installation
### Pre-built Binaries
Download from [GitHub Releases](https://github.com/catesandrew/mcp-agent-bridge/releases) for macOS (ARM64/x64), Linux (x64/ARM64), and Windows (x64).
### From Source
```bash
bun install
bun run build:exe # Standalone binaries in exe/
./install.sh # Copy to ~/.local/bin
```
### Background Services
```bash
# macOS -- LaunchAgents
./install.sh --launchd --load
# Windows -- NSSM services
.\examples\windows\install.ps1 -HttpService
# Linux -- systemd (see docs)
```
See the [Deployment Guide](https://catesandrew.github.io/mcp-agent-bridge/docs/deployment/macos) for details.
## Configuration
All servers are configured via environment variables:
| Variable | Default | Description |
|----------|---------|-------------|
| `CLAUDE_REVIEW_MODEL` | `opus` | Model for Claude reviews |
| `CLAUDE_MCP_HTTP_PORT` | `8940` | Claude HTTP port |
| `CODEX_MCP_HTTP_PORT` | `8941` | Codex HTTP port |
| `COPILOT_MCP_HTTP_PORT` | `8945` | Copilot HTTP port |
See [all environment variables](https://catesandrew.github.io/mcp-agent-bridge/docs/configuration/environment-variables) in the docs.
## Testing
```bash
bun run test # 36 tests across 5 suites
bun run lint # TypeScript type checking
```
## Security
- All proxies bind to `127.0.0.1` (localhost only)
- Read-only tool allowlist by default (`Read`, `Grep`, `Glob`, `LS`)
- Working directory validation against allowed roots
- Input size limits (500KB) and process timeouts (5 min)
## Prerequisites
- [Bun](https://bun.sh/) >= 1.0 (build)
- Node.js >= 20 (runtime)
- At least one agent CLI installed and authenticated:
[Claude](https://docs.anthropic.com/en/docs/claude-code) |
[Codex](https://github.com/openai/codex) |
[Copilot](https://githubnext.com/projects/copilot-cli)
- [`mcp-proxy`](https://github.com/nicholasgasior/mcp-proxy) (for HTTP mode)
## Contributing
See [CONTRIBUTING.md](CONTRIBUTING.md) for development setup and guidelines.
## License
Licensed under the Apache License, Version 2.0. See [LICENSE](LICENSE) for details.