A2A Client MCP Server
# A2A Client MCP Server
An MCP server that acts as a client to the Agent-to-Agent (A2A) protocol, allowing LLMs to interact with A2A agents through the Model Context Protocol (MCP).
## Features
- Connect to any A2A-compatible agent
- Send and receive messages
- Track and manage tasks
- Support for streaming responses
- Query agent capabilities and metadata
## Installation
```bash
# Install globally
npm install -g a2a-client-mcp-server
# Or run directly with npx
npx a2a-client-mcp-server
```
## Configuration
### Environment Variables
- `A2A_ENDPOINT_URL`: URL of the A2A agent to connect to (default: "http://localhost:41241")
## Usage with Claude Desktop
Add this to your `claude_desktop_config.json`:
### NPX
```bash
npm run build
npm link
```
```json
{
"mcpServers": {
"a2a-client": {
"command": "npx",
"args": ["-y", "a2a-client-mcp-server"],
"env": {
"A2A_ENDPOINT_URL": "http://localhost:41241"
}
}
}
}
```
### Docker
Build the Docker image:
```bash
docker build -t a2a-client-mcp-server .
```
Configure Claude Desktop:
```json
{
"mcpServers": {
"a2a-client": {
"command": "docker",
"args": [
"run",
"-i",
"--rm",
"-e",
"A2A_ENDPOINT_URL",
"a2a-client-mcp-server"
],
"env": {
"A2A_ENDPOINT_URL": "http://localhost:41241"
}
}
}
}
```
## Available Tools
### a2a_send_task
Send a task to an A2A agent
- `message` (string): Message to send to the agent
- `taskId` (string, optional): Task ID (generated if not provided)
### a2a_get_task
Get the current state of a task
- `taskId` (string): ID of the task to retrieve
### a2a_cancel_task
Cancel a running task
- `taskId` (string): ID of the task to cancel
### a2a_send_task_subscribe
Send a task and subscribe to updates (streaming)
- `message` (string): Message to send to the agent
- `taskId` (string, optional): Task ID (generated if not provided)
- `maxUpdates` (number, optional): Maximum updates to receive (default: 10)
### a2a_agent_info
Get information about the connected A2A agent
- No parameters required
## Resources
The server provides access to two MCP resources:
- `a2a://agent-card`: Information about the connected A2A agent
- `a2a://tasks`: List of recent A2A tasks
## Example Usage
This example shows how to use A2A Client MCP Server to interact with a Coder Agent:
```
First, let me explore what A2A agent we're connected to.
I'll use the a2a_agent_info tool to check the agent details.
The agent provides a coding service that can generate files based on natural language instructions. Let's create a simple Python script.
I'll use the a2a_send_task tool to send a request:
Task: "Create a Python function that calculates the Fibonacci sequence"
Now I can check the task status using a2a_get_task with the task ID from the previous response.
The agent has created the requested Python code. I can now retrieve and use this code in my project.
```
## Development
```bash
# Install dependencies
npm install
# Build the project
npm run build
# Run in development mode
npm run watch
```
## License
MIT
TDQS
Scored across 5 tools
Each tool has a clearly distinct purpose: a2a_agent_info retrieves agent metadata, a2a_send_task initiates a task, a2a_send_task_subscribe adds streaming, a2a_get_task checks status, and a2a_cancel_task stops execution. There is no overlap in functionality, making tool selection straightforward for an agent.
All tools follow a consistent 'a2a_verb_noun' pattern with snake_case, using descriptive verbs like 'get', 'send', and 'cancel'. This uniformity makes the tool set predictable and easy to navigate.
With 5 tools, the server is well-scoped for managing A2A agent tasks, covering core operations like sending, monitoring, and canceling tasks, plus agent info. This count is neither too sparse nor bloated, fitting the domain perfectly.
The tool set provides strong coverage for task lifecycle management (send, get, cancel) and agent discovery, with a2a_send_task_subscribe offering advanced streaming. A minor gap might be the lack of tools for listing or managing multiple tasks at once, but core workflows are well-supported.