agent-bridge

AgentBridge is a local, two-process bridge that enables bidirectional collaboration between Claude Code and Codex during the same working session by forwarding messages via an MCP channel (Claude side) and a local Codex app-server proxy/control WebSocket (Codex side).

Evaluated Mar 30, 2026 (0d ago)

Repo ↗ API Gateway agent-collaboration claude-code codex cli mcp local-first multi-agent developer-tools

⚙ Agent Friendliness

/ 100

Can an agent use this?

🔒 Security

/ 100

Is it safe for agents?

⚡ Reliability

/ 100

Does it work consistently?

Score Breakdown

⚙ Agent Friendliness

MCP Quality

Documentation

Error Messages

Auth Simplicity

Rate Limits

🔒 Security

TLS Enforcement

Auth Strength

Scope Granularity

Dep. Hygiene

Secret Handling

Runs as a local bridge using loopback ports and local processes; README does not document TLS or any auth controls, so it should not be treated as a hardened security boundary. It also explicitly notes no hardened boundary between untrusted tools.

⚡ Reliability

Uptime/SLA

Version Stability

Breaking Changes

Error Recovery

Best When

You trust both local agent runtimes (Claude Code and Codex) and want local, real-time message passing with a single shared working session.

Avoid When

You need strong security isolation, require multi-project concurrency, or expect a fully featured event stream beyond forwarded agentMessage items.

Use Cases

• Human-in-the-loop multi-agent collaboration across Claude Code and Codex in a single workspace
• Bidirectional coordination where one agent relays completed work to the other in real time
• Developer workflows that want both Claude Code and Codex available simultaneously

Not For

• A hosted or multi-tenant service
• A hardened security boundary between untrusted tools
• Generic orchestration for arbitrary agent backends
• Multi-session / multi-project support on the same machine (current limitation)

Interface

REST API

GraphQL

gRPC

MCP Server

Yes

SDK

Webhooks

Authentication

Methods: Local plugin channel (Claude Code plugin integration) Local daemon control WebSocket (no auth details documented)

OAuth: No Scopes: No

No authentication mechanism is described in the README. Communication appears to rely on local loopback ports and process lifecycle coordination.

Pricing

Free tier: No

Requires CC: No

Open-source local tool; any compute/costs come from running Claude Code and Codex models/tools, not from AgentBridge itself.

Agent Metadata

Pagination

none

Idempotent

True

Retry Guidance

Not documented

Known Gotchas

⚠ Fixed ports mean only one AgentBridge instance per machine (multi-project planned post-v1).
⚠ Codex sandbox blocks all writes to `.git` (commits/push/pull/checkout/merge may hang).
⚠ Only forwards `agentMessage` events, not intermediate events like `commandExecution` or `fileChange`.
⚠ Single Codex thread and single Claude foreground connection; new Claude sessions replace the previous one.

Alternatives

Use Claude Code and Codex separately with a shared repo and manual coordination Build a custom local adapter that bridges Claude Code MCP and Codex via their respective APIs Use other agent collaboration tooling that supports multi-agent orchestration with stronger isolation

Full Evaluation Report

Comprehensive deep-dive: security analysis, reliability audit, agent experience review, cost modeling, competitive positioning, and improvement roadmap for agent-bridge.

AI-powered analysis · PDF + markdown · Delivered within 30 minutes

$99

Package Brief

Quick verdict, integration guide, cost projections, gotchas with workarounds, and alternatives comparison.

Delivered within 10 minutes

Score Monitoring

Get alerted when this package's AF, security, or reliability scores change significantly. Stay ahead of regressions.

Continuous monitoring

$3/mo

API endpoint ↗ Agent guide ↗ Report inaccuracy

Scores are editorial opinions as of 2026-03-30.