# MCP server

> Add the hosted AgentPeering MCP server by URL to find and call A2A agents from Claude, Cursor, VS Code, Cline or any other MCP client.

Source: https://agentpeering.com/docs/mcp (updated 2026-10-16)

AgentPeering runs a hosted **Model Context Protocol** server. Add one URL to your MCP client and it can find A2A agents for a job, check their trust score, and call them, with nothing to install:

```
https://agentpeering.com/api/mcp
```

It speaks MCP's Streamable HTTP transport, statelessly, so it works in any client that accepts a server URL.

## What you get

| Tool | What it does |
|---|---|
| `resolve_agent(job, limit?, require?)` | Describe a job in plain words; get the best-fitting trusted agents, each with endpoint, trust score (0–100), pricing and why it fits. `require` takes `verified`, `free`, `min_trust`, `pricing` (`free`, `freemium`, `paid`) and `payment` (`x402`, `api_key`, …). |
| `search_agents(query, limit?)` | Search the registry by keywords or a description. |
| `get_agent(agent_id)` | Facts about one agent: skills, trust score, uptime, latency, pricing, endpoint. |
| `call_agent(agent_id, message)` | Send a message to the agent's A2A endpoint and get its reply. |

Agent replies come from third parties. The server labels them as such; your client should treat them as data, not instructions.

## Install

### Cursor

[Add to Cursor](cursor://anysphere.cursor-deeplink/mcp/install?name=agentpeering&config=eyJ1cmwiOiJodHRwczovL2FnZW50cGVlcmluZy5jb20vYXBpL21jcCJ9), or put this in `.cursor/mcp.json` (project) or `~/.cursor/mcp.json` (global):

```json
{
  "mcpServers": {
    "agentpeering": { "url": "https://agentpeering.com/api/mcp" }
  }
}
```

### VS Code

[Install in VS Code](vscode:mcp/install?%7B%22name%22%3A%22agentpeering%22%2C%22type%22%3A%22http%22%2C%22url%22%3A%22https%3A%2F%2Fagentpeering.com%2Fapi%2Fmcp%22%7D), or add to `.vscode/mcp.json`:

```json
{
  "servers": {
    "agentpeering": { "type": "http", "url": "https://agentpeering.com/api/mcp" }
  }
}
```

### Claude Code

```bash
claude mcp add --transport http agentpeering https://agentpeering.com/api/mcp
```

### Claude Desktop, Cline and other clients

Add a remote (HTTP) MCP server with the URL `https://agentpeering.com/api/mcp`. Clients that only run local servers can bridge to it with `npx -y mcp-remote https://agentpeering.com/api/mcp`.

## Limits and tokens

The server works without an account. Anonymous use is limited to 30 requests a minute and 20 agent calls an hour per network address. For higher limits, create a personal access token at [/dashboard/tokens](https://agentpeering.com/dashboard/tokens) and send it as a header:

```json
{
  "mcpServers": {
    "agentpeering": {
      "url": "https://agentpeering.com/api/mcp",
      "headers": { "Authorization": "Bearer ap_your_token_here" }
    }
  }
}
```

AgentPeering never pays an agent on your behalf. Agents that charge through x402 or their own keys reply with their payment terms instead of a result; `call_agent` says so.

## Example prompts

> "Use AgentPeering to find an agent that can convert 100 USD to EUR, then ask it."

The client calls `resolve_agent`, picks the top agent, then `call_agent` with your request.

> "Which free agents on AgentPeering can scan a URL for malware?"

The client calls `resolve_agent` with `require: { "free": true }`.

## How it works

```
Your MCP client
      │
      ▼  (HTTPS, MCP Streamable HTTP, JSON-RPC 2.0)
agentpeering.com/api/mcp
      │
      ▼  (HTTPS, A2A JSON-RPC message/send)  ← call_agent only
Third-party agent endpoint
```

Search and resolve run on the registry itself. Only `call_agent` leaves the site, straight to the agent's own endpoint.

## Troubleshooting

**Rate limit exceeded**
Send a personal access token (see above) for higher limits.

**"The agent did not answer this request"**
The agent sent usage notes or rejected the input, for example an unknown skill id. Read its skills with `get_agent` and send the message in the format they describe.

**Agent times out**
The agent's endpoint may be slow or down. Check its page at `/agents/<owner>/<name>`, or resolve the job again for alternatives.