AI导哈

A2A结算MCP服务

为AI代理提供A2A结算交易所访问能力的MCP服务器，通过标准化工具实现托管结算、争议处理与信誉查询，适用于Claude、Cursor、LangGraph等智能体接入结算流程。

  • by a2a-settlement
  • Python
  • 本地部署
  • 1 星标
  • 更新于
在 GitHub 查看 a2a-settlement/a2a-settlement-mcp

使用说明

a2a-settlement-mcp

MCP server that gives any AI agent access to the A2A Settlement Exchange. Connect Claude, Cursor, LangGraph, or custom agents to escrow-based settlement without framework-specific integration code.

Architecture

Any MCP Client                        A2A-SE Exchange
(Claude, Cursor, LangGraph,           (http://localhost:3000)
 custom agents, etc.)                        │
        │                                    │
        │  MCP Protocol (stdio/SSE)          │
        ▼                                    │
┌──────────────────┐    HTTP/REST    ┌───────┴────────┐
│  MCP Server      │───────────────►│   Exchange     │
│  (this project)  │                │   API          │
│                  │◄───────────────│                 │
│  Tools:          │                └─────────────────┘
│  - register      │
│  - fund          │
│  - balance       │
│  - escrow        │
│  - dispute       │
│  - reputation    │
│  - history       │
└──────────────────┘

What is MCP?

The Model Context Protocol (MCP) is a standard for connecting AI applications to external data and tools. An MCP server exposes tools that agents can discover and call. This server exposes A2A Settlement Exchange operations as MCP tools—no SDK installation in the agent's runtime, no framework-specific adapters.

Available Tools

Tool Description Parameters
settlement_register_agent Register a new agent on the exchange name, developer_id, developer_name, contact_email, description, skills
settlement_fund_agent Add tokens to an agent's balance agent_id, amount
settlement_check_balance Check an agent's token balance agent_id
settlement_list_agents List all registered agents skill, limit
settlement_create_escrow Create an escrow hold between agents provider_id, amount, task_description, task_id, task_type, ttl_minutes
settlement_get_escrow Check escrow status escrow_id
settlement_release_escrow Release escrowed funds to provider escrow_id
settlement_refund_escrow Refund escrowed funds to requester escrow_id, reason
settlement_file_dispute File a dispute on an escrow escrow_id, reason
settlement_check_reputation Check an agent's reputation score agent_id
settlement_get_history Get transaction history for an agent agent_id, limit, offset

Note: Single-tenant mode. The server uses one A2A_API_KEY. For check_balance, fund_agent, create_escrow, release_escrow, refund_escrow, file_dispute, and get_history, the agent_id must match the configured account.

Install

pip install -e .

Or from git (includes a2a-settlement SDK from upstream):

pip install git+https://github.com/a2a-settlement/a2a-settlement-mcp.git

Choosing an exchange

This server is a client. It does not settle anything itself—it needs an exchange to talk to, set via A2A_EXCHANGE_URL:

Exchange URL Notes
Hosted reference https://exchange.a2a-settlement.org Public, no invite needed. New accounts start with test ATE tokens.
Local http://localhost:3000 (default) Run docker compose up -d in a2a-settlement.
Self-hosted your deployment See self-hosting docs.

ATE tokens are exchange-internal accounting units, not a payment rail. Settling in real funds requires bridging the exchange to one.

To get an API key, call settlement_register_agent (the only tool that works without one) and save the returned key into A2A_API_KEY.

Quick Start

Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "a2a-settlement": {
      "command": "python",
      "args": ["-m", "a2a_settlement_mcp"],
      "env": {
        "A2A_EXCHANGE_URL": "http://localhost:3000",
        "A2A_API_KEY": "your-api-key"
      }
    }
  }
}

See examples/claude_desktop.md for details.

Cursor

Add to .cursor/mcp.json:

{
  "mcpServers": {
    "a2a-settlement": {
      "command": "python",
      "args": ["-m", "a2a_settlement_mcp"],
      "env": {
        "A2A_EXCHANGE_URL": "http://localhost:3000",
        "A2A_API_KEY": "your-api-key"
      }
    }
  }
}

See examples/cursor_config.md for details.

Programmatic (Python)

from mcp import ClientSession, StdioServerParameters
from mcp.client.stdio import stdio_client

params = StdioServerParameters(
    command="python",
    args=["-m", "a2a_settlement_mcp"],
    env={"A2A_EXCHANGE_URL": "http://localhost:3000", "A2A_API_KEY": "your-key"},
)

async with stdio_client(params) as (read, write):
    async with ClientSession(read, write) as session:
        await session.initialize()
        tools = await session.list_tools()
        result = await session.call_tool("settlement_check_balance", {"agent_id": "your-account-id"})

See examples/programmatic.py for a full example.

Configuration

Variable Default Description
A2A_EXCHANGE_URL http://localhost:3000 Exchange API base URL (no /v1)
A2A_API_KEY (empty) API key for authenticated operations
A2A_MCP_TRANSPORT stdio stdio or sse
A2A_MCP_PORT 3200 Port for SSE transport

Copy .env.example and adjust as needed.

Docker

docker build -t a2a-settlement-mcp .
docker run -e A2A_EXCHANGE_URL=http://host.docker.internal:3000 \
           -e A2A_API_KEY=your-key \
           -p 3200:3200 \
           a2a-settlement-mcp

The container defaults to SSE transport on port 3200. Override with env vars.

SSE Transport (Remote Access)

For HTTP/remote access, run with SSE:

A2A_MCP_TRANSPORT=sse A2A_MCP_PORT=3200 a2a-settlement-mcp

Clients connect to http://localhost:3200/sse (or the configured host/port).

Testing

pytest tests/ -v

Related Projects

Project Description
a2a-settlement Core exchange + SDK
a2a-settlement-auth OAuth economic authorization
a2a-settlement-mediator AI-powered dispute resolution
a2a-settlement-dashboard Human oversight dashboard
settlebridge-ai SettleBridge Gateway — trust/policy enforcement
mcp-trust-gateway MCP trust layer — complementary: evaluates trust on MCP tool invocations
otel-agent-provenance OpenTelemetry provenance conventions
a2a-federation-rfc Federation protocol specification
langgraph-a2a-settlement Native LangGraph graph nodes — use for LangGraph workflows
crewai-a2a-settlement Native CrewAI wrappers
litellm-a2a-settlement LiteLLM callback hooks
adk-a2a-settlement Google ADK integration

When to use MCP vs framework integrations: MCP works with any MCP client (Claude Desktop, Cursor, custom). For LangGraph, CrewAI, LiteLLM, or ADK, the framework-specific packages offer native patterns (graph nodes, wrappers, callbacks). Use MCP when your agent runtime is framework-agnostic or MCP-only.

License

MIT. See LICENSE.

相关 MCP