Cloudflare Agents SDK

The Cloudflare Agents SDK (agents) connects to Edison with its built-in MCP client. Each agent instance is a Durable Object with its own storage, so this.mcp is a long-lived MCPClientManager rather than a client you build per request. Your connection URL carries your API key, so no auth header is needed - and because Edison authenticates from the URL path, you never wire up the SDK's OAuth flow for this connection.

npm install agents ai @ai-sdk/openai

Connect in onStart() and hand the tools to your model. onStart() runs after the SDK has restored any existing MCP connections, so it's the right place for this:

import { Agent, type AgentNamespace, routeAgentRequest } from 'agents';
import { generateText } from 'ai';
import { openai } from '@ai-sdk/openai';

interface Env {
  EdisonAgent: AgentNamespace<EdisonAgent>;
  EDISON_MCP_URL: string;
}

export class EdisonAgent extends Agent<Env> {
  async onStart() {
    await this.addMcpServer('edison', this.env.EDISON_MCP_URL, {
      transport: {
        type: 'streamable-http',
        // See "Keep a stable session across turns" below - this is not optional.
        headers: { 'x-edison-conversation-id': this.name },
      },
    });
  }

  async onRequest(_request: Request) {
    const { text } = await generateText({
      model: openai('gpt-5.6-luna'),
      tools: this.mcp.getAITools(),
      messages: [{ role: 'user', content: 'List my available tools.' }],
    });
    return Response.json({ text });
  }
}

export default {
  async fetch(request: Request, env: Env) {
    return (await routeAgentRequest(request, env)) ?? new Response('Not found', { status: 404 });
  },
};

Set EDISON_MCP_URL to your connection URL, e.g. https://mcp.edison.watch/mcp/<your-api-key>/?client=cloudflare-agents. Keep it in a Worker secret (wrangler secret put EDISON_MCP_URL), not in wrangler.jsonc - the URL is the credential. Agents also need a Durable Object binding and migration entry in your Wrangler config; the agents-starter template ships both.

Keep a stable session across turns to preserve data-leak protection

Send a stable x-edison-conversation-id header on every turn of the same conversation. That header is what keeps Edison's data-leak protection intact across a multi-turn run: Edison tracks lethal-trifecta risk per session, so if each turn looks like a brand-new session, that protection resets - and a later turn can leak data that the accumulated risk should have blocked.

This SDK makes that unusually easy, which is why it's already wired into the snippet above. You normally address one agent instance per conversation - getAgentByName(env.EdisonAgent, conversationId), or the /agents/edison-agent/<conversationId> route that routeAgentRequest resolves - so the agent's own Durable Object name, this.name, is already a stable per-conversation identifier that survives hibernation. Passing it straight through is all it takes.

Optional: the encrypted-secrets header

For servers with zero-knowledge-encrypted secrets, add it alongside the conversation id under transport.headers:

await this.addMcpServer('edison', this.env.EDISON_MCP_URL, {
  transport: {
    type: 'streamable-http',
    headers: {
      'x-edison-conversation-id': this.name,
      'x-edison-secret-key': this.env.EDISON_SECRET_KEY,
    },
  },
});

Governing the other direction

An Agents SDK agent can also expose MCP tools (as an McpAgent), which puts it on the server side of the gateway - Edison can then front your agent's own tools for other clients. That's the Add an MCP server path, and it's independent of everything above. Cloudflare's own hosted server (mcp.cloudflare.com/mcp, for Workers and DNS management) is available in Edison's server marketplace.