Python: Integration: cowork-to-code-bridge for local Claude Code execution in agents

@sphenry is already working on this.

Since Jun 17, 2026.

## Problem

Agent Framework users building enterprise agents need reliable local execution for code-heavy tasks without:
- Separate API key management
- Cloud-only execution models
- Additional billing outside existing subscriptions

## Solution: cowork-to-code-bridge MCP Provider

**cowork-to-code-bridge** provides exactly that: agents escalate to Claude Code on the user's infrastructure using their existing subscription.

### Key Features

- ✅ **No API keys** — uses Claude subscription on the user's machine
- ✅ **Full local context** — agent has access to repos, shell, MCPs, environment
- ✅ **MCP standard** — JSONRPC 2.0 over stdio (native protocol)
- ✅ **Zero network ports** — file-based queue, token-authenticated
- ✅ **Production-ready** — 570 lines + 9 unit tests

### How It Works

**Agent Framework config:**
```json
{
"providers": {
"claude-code-bridge": {
"type": "mcp",
"command": "cowork-to-code-bridge-mcp",
"args": ["--stdio"],
"env": {"BRIDGE_ROOT": "$HOME/.cowork-to-code-bridge"}
}
}
}
```

**Agent escalates work:**
```python
result = agent.escalate(
tool="escalate_to_claude",
request="Debug the API failure and propose fix",
wait_seconds=600
)
```

## Why This Matters

Agent Framework users building enterprise agents need reliable local execution for code-heavy tasks. cowork-to-code-bridge provides that without API key management: agents escalate to Claude Code on the user's infrastructure using their existing subscription.

Cost-effective, local-first alternative to cloud APIs.

## Reference Implementation

- **GitHub:** https://github.com/abhinaykrupa/cowork-to-code-bridge
- **MCP Docs:** https://github.com/abhinaykrupa/cowork-to-code-bridge/blob/main/docs/MCP_SERVER_IMPLEMENTATION.md
- **Live examples:** Crew.ai, Hermes, Open Claw

## Proposed Action

Options:
1. Integration example showing MCP bridge provider in Agent Framework config
2. Use case: "Build → test → debug workflow with local Claude Code execution"
3. Or link from docs to our starter repo

(Happy to contribute any of these.)