SDK guides

Connectors is a remote HTTP MCP server. Every framework below loads it the same way — point at CONNECTORS_MCP_URL and pass the bearer token.

Claude Agent SDK

Use the MCP server as an HTTP MCP server in the SDK options.

import { query } from "@anthropic-ai/claude-agent-sdk";

const mcpServers = {
  connectors: {
    type: "http" as const,
    url: process.env.CONNECTORS_MCP_URL!,
    headers: {
      Authorization: `Bearer ${process.env.CONNECTORS_API_KEY}`,
    },
  },
};

for await (const message of query({
  prompt: "Find the latest Linear issues about onboarding and summarize the top blockers.",
  options: { mcpServers, maxTurns: 20 },
})) {
  if (message.type === "result") {
    console.log(message);
  }
}

OpenAI SDK

Use the Responses API with an MCP tool. For authenticated remote MCP servers, pass the token in the MCP tool’s authorization field.

import OpenAI from "openai";

const client = new OpenAI();

const response = await client.responses.create({
  model: process.env.OPENAI_MODEL!, // any model your account has enabled for the Responses API
  input: "Search Connectors for recent Linear issues about onboarding and summarize the blockers.",
  tools: [
    {
      type: "mcp",
      server_label: "canvas_connectors",
      server_description: "Canvas Connectors workspace search and app context.",
      server_url: process.env.CONNECTORS_MCP_URL!,
      authorization: process.env.CONNECTORS_API_KEY!,
      require_approval: "never",
      allowed_tools: [
        "sketch_search",
        "sketch_search_entities",
        "sketch_get_entity_context",
        "sketch_get_file_content",
      ],
    },
  ],
});

console.log(response.output_text);

For production, choose require_approval based on the sensitivity of the connected data and action tools. The public Sketch MCP tools are read-oriented today, but live app-action tools should generally require approval unless the workflow is narrowly scoped.

LangChain

Use @langchain/mcp-adapters to load Connectors tools into a LangChain agent.

import { MultiServerMCPClient } from "@langchain/mcp-adapters";
import { createAgent } from "langchain";
import { ChatOpenAI } from "@langchain/openai";

const mcpClient = new MultiServerMCPClient({
  canvasConnectors: {
    transport: "http",
    url: process.env.CONNECTORS_MCP_URL!,
    headers: {
      Authorization: `Bearer ${process.env.CONNECTORS_API_KEY}`,
    },
  },
});

const tools = await mcpClient.getTools();

const agent = createAgent({
  model: new ChatOpenAI({ model: process.env.OPENAI_MODEL! }), // any model your account has enabled
  tools,
});

const result = await agent.invoke({
  messages: [
    { role: "user", content: "Find the Linear issue for the new onboarding flow and draft a Slack summary." },
  ],
});

console.log(result);

Raw HTTP

List tools:

const listResponse = await fetch(process.env.CONNECTORS_MCP_URL!, {
  method: "POST",
  headers: {
    Authorization: `Bearer ${process.env.CONNECTORS_API_KEY}`,
    Accept: "application/json, text/event-stream",
    "Content-Type": "application/json",
    "Mcp-Protocol-Version": "2025-03-26",
  },
  body: JSON.stringify({ jsonrpc: "2.0", id: 1, method: "tools/list" }),
});

console.log(await listResponse.json());

Call search:

const callResponse = await fetch(process.env.CONNECTORS_MCP_URL!, {
  method: "POST",
  headers: {
    Authorization: `Bearer ${process.env.CONNECTORS_API_KEY}`,
    Accept: "application/json, text/event-stream",
    "Content-Type": "application/json",
    "Mcp-Protocol-Version": "2025-03-26",
  },
  body: JSON.stringify({
    jsonrpc: "2.0",
    id: 2,
    method: "tools/call",
    params: {
      name: "sketch_search",
      arguments: { query: "customer onboarding launch blockers", kind: "task", sortBy: "recency", limit: 5 },
    },
  }),
});

console.log(await callResponse.json());