Your agent is only as useful as the tools it can reach. If those tools live behind a remote MCP server like Linear, GitHub or an internal service, getting there has historically meant writing your own client and your own auth plumbing.

@loopstack/mcp-module ships two workflow tools that connect any Loopstack agent to a remote Model Context Protocol  server over Streamable HTTP or legacy SSE. No transport code, no header juggling.

Two Tools, One Job

• McpListToolsTool — discover what a remote MCP server exposes.
• McpCallTool — invoke one of those tools.

Both take a serverUrl per call, so a single agent can hop between every host on its allowlist within the same conversation. Switch hosts mid-conversation without re-registering anything.

Inject It Like Any Other Tool

The whole module follows the same @InjectTool pattern you already use everywhere else:

@InjectTool({
  allowedHosts: ['mcp.linear.app'],
  hostHeaderEnv: {
    'mcp.linear.app': { Authorization: 'LINEAR_MCP_TOKEN' },
  },
})
private mcpCallTool: McpCallTool;

Two lines of config and your agent is talking to Linear. Add another entry to allowedHosts and another hostHeaderEnv mapping and it’s talking to GitHub too. Auth tokens are sourced from process.env at call time and never written into static config or logged. Also an allowlist plus DNS resolution check keeps a model-controlled URL from wandering into your internal network.

Try It

npm install @loopstack/mcp-linear-example-workflow

Or to copy the source into your project so you can modify it:

npx giget@latest gh:loopstack-ai/loopstack/registry/examples/mcp-linear-example-workflow#main

Follow the README and you’ll get a working chat agent wired to Linear’s hosted MCP server. Set LINEAR_MCP_TOKEN="Bearer lin_oauth_..." in your env and you’re done. Point a workflow at it and your agent can list and create issues from chat.

For the full security model, transport selection, and the env-reader / metrics extension points, the @loopstack/mcp-module README walks through every knob.

Wrapping Up

MCP is becoming the integration layer for AI tools, and the most valuable agents are the ones that can reach beyond a single vendor. This module makes that reach trivial to configure. We’re excited to see what you connect to it.

— Your Loopstack Team

Ready to join the conversation? Check out our GitHub , connect with us on Discord , and follow our journey on LinkedIn .