First release of a2a-mcp-bridge: two Claude agents collaborating on one machine, with MCP for tools and A2A for agent-to-agent delegation, running on Claude on Vertex AI (Anthropic API fallback; offline mock for tests/CI).
Highlights
- Researcher agent (MCP tools:
web_search,fetch_document) delegates to an Analyst agent over A2A (MCP tools:summarize,score_relevance,extract_entities). - One-command demo that runs with no credentials via a deterministic offline provider.
- MCP façade (
research_and_analyze) exposing the whole system as a single tool — usable as a skill in Claude Code, Codex, and opencode (seeintegrations/). - Provider abstraction: Vertex AI → Anthropic API → mock.
- Offline test suite (providers, MCP servers, A2A round-trip, end-to-end, façade) green on Python 3.10–3.12.
Install
pip install -e ".[dev]"
a2a-mcp-bridge "the impact of remote work on team productivity"See the README and docs/ARCHITECTURE.md.