Skip to main content

Agent as MCP Endpoint

Any RickyData agent can be exposed as an MCP server, letting MCP clients call agent capabilities as individual tools.

Concept​

When you connect to an agent via MCP, the agent's capabilities are listed as tools. This means Claude Code, Cursor, or any MCP client can call agent-specific operations without going through the chat interface.

SDK usage​

import { AgentMCPClient } from 'rickydata';

const client = new AgentMCPClient({ token: 'mcpwt_...' });

// Connect to an agent (MCP initialize handshake)
const info = await client.connect('research-agent');
console.log(info.name, info.version);

// List the agent's tools
const tools = await client.listTools('research-agent');
tools.forEach(t => console.log(t.name, '-', t.description));

// Call a tool
const result = await client.callTool(
  'research-agent',
  'web_research',
  { topic: 'MCP protocol updates' }
);
console.log(result);

CLI usage​

# List an agent's MCP tools
rickydata mcp agent tools research-agent

# Call an MCP tool on an agent
rickydata mcp agent call research-agent web_research \
  '{"topic":"MCP protocol updates"}'

Claude Code integration​

Register an agent as an MCP server in Claude Code:

claude mcp add --transport http research-agent \
  https://agents.rickydata.org/agents/research-agent/mcp \
  --header "Authorization:Bearer mcpwt_..."

After adding, Claude Code can discover and use the agent's tools directly.

How it works​

The Agent Gateway exposes each agent at a per-agent MCP endpoint:

POST /agents/:agentId/mcp

This endpoint:

  1. Validates wallet authentication
  2. Returns the agent's tool definitions via tools/list
  3. Routes tools/call through the agent's execution pipeline
  4. The agent can use MCP Gateway tools during execution (tools calling tools)

Use cases​

  • Research agent as a tool — let Claude Code call web_research, paper_search, or summarize without managing a full chat session
  • Coding agent as a reviewer — call review_code or suggest_fixes as standalone operations
  • Chaining agents — one agent calls another agent's MCP tools through the gateway
  • Canvas workflows — agent nodes in canvas DAGs use the MCP interface internally

Next steps​