codex-bridge
MCP server that lets Claude Code ask GPT Codex for adversarial planning, code review, debugging, research, and risk triage without leaving your project workflow.
README
codex-bridge
Give Claude Code a Codex sparring partner.
codex-bridge is an MCP server that lets Claude Code ask GPT Codex for adversarial planning, code review, debugging, research, and risk triage without leaving your project workflow.
The npm package is @ndcorder/codex-bridge. The installed CLI binary and MCP server key are both codex-bridge.
What It Does
codex_plan: get an implementation plan with trade-offs and stop-ship checks before writing code.codex_review: ask Codex for critical code review, with evidence mode on by default.codex_ask: brainstorm, sanity-check assumptions, or compare approaches.codex_risk_radar: score a diff and choose the right collaboration depth.codex_research: run multi-turn research and write durable research artifacts.codex_debug: run structured root-cause analysis for hard bugs.codex_stats,codex_sessions,codex_reset: inspect and manage bridge sessions.
Requirements
- Node.js 20 or newer is recommended. CI runs on Node 20 and 22.
- The Codex CLI must be installed, authenticated, and available as
codexonPATH. - An MCP-capable client such as Claude Code.
Check the Codex CLI before configuring the bridge:
codex --version
Quick Start
From the project where you want Claude Code to use Codex:
npx -y @ndcorder/codex-bridge init
This command:
- creates or updates
.mcp.jsonwith acodex-bridgestdio server entry - preserves existing MCP servers
- appends Codex Bridge guidance to
CLAUDE.mdif it is not already present
Restart Claude Code after running the init command.
Manual MCP Configuration
If you prefer to edit MCP config yourself, add this server entry:
{
"mcpServers": {
"codex-bridge": {
"type": "stdio",
"command": "npx",
"args": ["-y", "@ndcorder/codex-bridge"]
}
}
}
First Workflows
Ask Codex to plan before a risky change:
Use codex_plan to review this approach before implementation:
<describe the feature, constraints, and files involved>
Ask for evidence-based review after changes:
Use codex_review with include_diff=true and evidence_mode=true.
Focus on security, data loss, and API regressions.
Use risk radar when deciding how much review a diff needs:
Use codex_risk_radar on the current diff and recommend whether to ask,
review, or plan then review.
See docs/USAGE.md for more workflows.
Configuration
Environment variables:
CODEX_BRIDGE_MODELorCODEX_MODEL: default model override. Default:gpt-5.3-codex.CODEX_BRIDGE_RETRIES: transient retry count. Default:1.CODEX_BRIDGE_RETRY_BACKOFF_MS: retry backoff base in milliseconds. Default:500.CODEX_BRIDGE_MAX_SESSIONS: max in-memory sessions before oldest-session eviction. Default:200.CODEX_BRIDGE_SESSION_TTL_MS: session inactivity TTL in milliseconds. Default:86400000.
Common per-request runtime options:
working_dir: project directory for Codex file access and implicit session key.timeout: timeout in milliseconds, max600000.session_id: explicit context isolation key.model: per-request Codex model override.retries: per-request retry count, max10.retry_backoff_ms: per-request retry backoff base, max60000.
Full tool schemas are in docs/TOOLS.md.
Documentation
docs/INSTALL.md: installation, MCP setup, and verification.docs/USAGE.md: practical workflows and examples.docs/TOOLS.md: complete MCP tool reference.docs/TROUBLESHOOTING.md: common setup and runtime failures.docs/ARCHITECTURE.md: maintainer architecture notes.CONTRIBUTING.md: development workflow.RELEASE.md: release checklist.SECURITY.md: vulnerability reporting.CHANGELOG.md: release history.
Local Development
npm install
npm run build
npm test
Run the full verification suite:
npm run verify
The full suite includes unit tests, the MCP stdio smoke test, and the bake-off black-box suite.
Release Commands
npm run release:bump: bump patch version inpackage.jsonandpackage-lock.json.npm run release:bump:patch|minor|major: explicit semver bump.npm run release:notes: generate.release/RELEASE_NOTES.mdfromCHANGELOG.md.npm run release:pack-check: build and runnpm pack --dry-run.npm run release:prepare: run verify, pack check, and release notes generation.
Publish scoped packages publicly with:
npm publish --access public
Package Contents
The npm package intentionally ships only runtime output and documentation:
dist/srcREADME.mdCHANGELOG.mddocs/*.md
Generated research artifacts are written to the consuming project's .codex-bridge/ directory when codex_research receives working_dir.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.