Codex Mcp Bridge
Wire Codex CLI into MCP so your IDE agent can query, review, search, assess, and fetch structured output with session control.
Overview
Codex MCP Bridge is an MCP server for the Build phase that exposes the Codex CLI as query, review, search, assess, structured, and session tools with configurable concurrency.
What is this MCP server?
- MCP tools: query, review, search, assess, structured, and session management for Codex CLI
- stdio npm package codex-mcp-bridge (v0.5.1)
- CODEX_MAX_CONCURRENT caps parallel Codex subprocesses (default 3)
- CODEX_MCP_SERVERS controls which ~/.codex/config.toml MCP servers stay enabled in subprocess
- CODEX_FALLBACK_MODEL for quota errors alongside CODEX_DEFAULT_MODEL
- Package version 0.5.1
- CODEX_MAX_CONCURRENT default: 3
- Six capability areas documented: query, review, search, assess, structured, sessions
Community signal: 3 GitHub stars.
What problem does it solve?
Orchestrator agents cannot safely drive the Codex CLI with review and assess modes unless you maintain fragile subprocess wrappers yourself.
Who is it for?
Builders on the Codex CLI who want MCP-native query, review, and structured tools inside a different IDE or multi-agent workflow.
Skip if: Teams not using Codex locally or workflows that need only a single IDE with no MCP delegation layer.
What do I get? / Deliverables
stdio MCP tools invoke Codex with session awareness, model fallbacks, and bounded concurrency so your primary agent delegates coding tasks cleanly.
- MCP tools for Codex query, review, search, assess, structured output, and sessions
- Concurrency-limited Codex subprocess pool (default max 3)
- Model fallback path when primary Codex model hits quota errors
Recommended MCP Servers
Journey fit
Codex CLI exposure through MCP is build-phase agent infrastructure for teams standardizing on OpenAI Codex alongside other hosts. The npm stdio bridge with concurrency limits and MCP server passthrough belongs in agent-tooling, not ship-time monitoring.
How it compares
Codex CLI MCP bridge with quota fallback and subprocess MCP filtering—not a hosted OpenAI API wrapper.
Common Questions / FAQ
Who is Codex MCP Bridge for?
Solo developers and small teams that run the Codex CLI daily and need another MCP host to call Codex with review, assess, and structured modes.
When should I use Codex MCP Bridge?
Use it while setting up build-phase agent tooling when your orchestrator must share Codex sessions and respect concurrency and nested MCP server rules.
How do I add Codex MCP Bridge to my agent?
Install codex-mcp-bridge via npm, register it as a stdio MCP server, set CODEX_CLI_PATH if needed, and tune CODEX_MAX_CONCURRENT, CODEX_DEFAULT_MODEL, CODEX_FALLBACK_MODEL, and CODEX_MCP_SERVERS as required.