a2a_and_mcp_talking_quantum

# A2A Talking Quantum – Minimal Agent-to-Agent (A2A) Demo This repository accompanies our Medium article **“A New Hope: When a SmolAgent Meets a LangGraph Agent to Talk About Quantum Squirrels”**. It demonstrates how to integrate (remote) agents and tools into a SmolAgents-based host via the A2A and MCP protocol, respectively. --- ## Highlights This example shows: - A **remote (or local) LangGraph-based literary critic agent**, wrapped in an A2A-compatible server - A **local SmolAgent host system**, able to reason and call both: - remote (or local) classic tools via **MCP** (e.g., library document tools), and - remote (or local) agents via **A2A** - A **proxy bootstrapping system** and **A2A Client** to expose remote A2A agents to the local SmolAgent host Agents use protocols to collaborate: **MCP for tools**, **A2A for peer agents**. --- ## Setup Create a new `python==3.12` environment. Activate it. ```bash git clone <repo-url> cd <repo-dir> pip install -r requirements.txt ``` If you are using `conda`, don't forget `conda install nb_conda_kernels` --- ## Optionally: Start the Remote Literary Critic (A2A Agent Server) This is the LangGraph-powered A2A agent server. Start it from the host notebook (below) or manually: ```bash python literature_critic.py --url <url> --openai-key <your openai key> ``` where `url` is e.g. "http://localhost:52042" (default) - Described by an `AgentCard` and accessible via A2A protocol --- ## Optionally: MCP Tooling (Library Server) In parallel, we also run a classic MCP tool server. Start it from the host notebook (below) or manually: ```bash python library_server.py ``` - Runs on port: `58002` (our default for the MCP server) - Serves document tools: `get_library`, `get_document` ## Starting the host The `host.ipynb` notebook brings together remote MCP tools and remote A2A agents with a local host to enable hybrid reasoning. To run it, simply open the Jupyter notebook and execute the cells in order. ```bash jupyter notebook host.ipynb ``` If you started the A2A and / or tool server on different, remote machines (not localhost), then adjust the urls in the notebook accordingly. --- ## Bootstrap the Proxy Agents The script `a2a_bootstrap_proxies.py` contains methods for discovering and registering remote A2A agents as SmolAgent-compatible proxies. It retrieves their `AgentCard`s and wraps each one in a lightweight proxy (`A2AMultiStepAgent`) that integrates smoothly with the SmolAgents host. --- ## The A2A Client The `a2a_client.py` script is used to construct and provide `A2AMultiStepAgent` — a class derived from SmolAgents' `MultiStepAgent` — and a client to communicate directly with a remote A2A agent. --- ## Core Files - `host.ipynb`: Sets up a compact local host and brings together the remote MCP tools and remote A2A agents - `library_server.py`: A2AStarletteApplication exposing the LangGraph agent - `literature_critic.py`: Defines the LangGraph-based literary critic agent - `a2a_client.py`: Sends A2A messages to the remote agent - `a2a_bootstrap_proxies.py`: Discovers remote agents and wraps them as proxy agents - `build_mcp_tool_schema.py`: Creates tool schemas for MCP tools - `library/`: Directory for document resources - `requirements.txt`: Required packages --- ## License Creative Commons BY-NC 4.0 For non-commercial, educational use. Dependencies retain their respective licenses. --- ## Contact 📧 [kontakt@seasparks.de](mailto:kontakt@seasparks.de)