Desktop app MCP server

The axiom.ai desktop app ships with a built-in MCP server. Once it's switched on, any MCP-aware LLM client running on the same machine can call your axiom.ai API: trigger automations, drive a cloud browser through step functions, check remaining runtime, and stop runs. The server speaks the standard Model Context Protocol, so Claude Desktop, Claude Code, Cursor, and every other MCP-compatible client work with it out of the box.

This is the easiest way to put axiom.ai inside an LLM. Skip ahead to Before you begin to enable it.

What it gives you

With the desktop app's MCP server enabled, an LLM client can:

  • Trigger automations by name. Ask Claude to run "Scrape today's leads", and it calls /trigger under the hood.
  • Pass input data. The LLM can format rows of input and send them in the same call.
  • Check run status. The LLM polls /run-data and reports back when the run finishes.
  • Drive a cloud browser step by step. The LLM can call step functions (goto, click, enterText, scrape, wait) directly, choosing each action based on what the previous one returned.
  • Read remaining runtime before kicking off expensive runs.
  • Stop runs if a conversation needs to abort one early.

No CLI, no SDK installation, no MCP server to write yourself. The desktop app handles the protocol; the LLM handles the conversation.

Before you begin

You need:

  • A paid axiom.ai account. API access is gated to paid plans. See the pricing page for plans that include the API.
  • The axiom.ai desktop app, installed and running on the same machine as your LLM client. Download from the desktop app page.
  • An axiom.ai API key, generated from the Dashboard. See Authentication.

Enable the MCP server

  1. Open the axiom.ai desktop app.
  2. Open the Claude MCP server panel from the desktop app's settings.
  3. Toggle the server on.
  4. Paste your axiom.ai API key into the API key field. The key is stored locally on your machine; the desktop app never sends it anywhere else.
  5. Confirm the panel reads Connected. Claude (or another MCP client) can now reach the server.

To rotate the key later, click Remove and reset and paste a new key. To turn the server off entirely, toggle it back off.

Use it from your LLM client

The built-in server speaks standard MCP, so any MCP-aware client can talk to it. The most common clients today:

  • Claude Desktop. Anthropic's official desktop app for the Claude chat experience. Once the axiom.ai desktop app's MCP server is on, Claude Desktop discovers it as a tool source. See Register with Claude for the connection details.
  • Claude Code. Anthropic's terminal-native coding agent. Run claude in a project and the agent picks up the MCP server automatically.
  • Cursor. Cursor's chat panel speaks MCP and treats the axiom.ai server like any other tool source.
  • Any other MCP client. The protocol is open, so Cline, Cowork, Continue, Zed's AI panel, and others all work the same way.

You don't need to install anything client-side beyond the LLM client itself. Once both the axiom.ai desktop app and the LLM client are running, the LLM can see and call axiom.ai tools.

Example conversations

A few prompts to try once the server is enabled:

"Run my 'Daily competitor prices' automation and tell me when it's done."

"Open a browser session, go to news.ycombinator.com, scrape the front-page titles, and summarise the top three."

"How much axiom runtime do I have left this month?"

Claude picks the right MCP tool for each task: /trigger for the first, a sequence of step functions for the second, and /remaining-runtime for the third.

What's under the hood

The built-in server is implemented with Anthropic's official TypeScript MCP SDK and embedded in the axiom.ai desktop app. The same SDK powers Claude Desktop's own tool integrations, so the server behaves identically to any other MCP server an LLM might encounter.

If you want to extend or customise the tool surface, see Build your own (TypeScript) for a standalone server you can fork. The built-in server is the recommended path for almost everyone; the build-your-own pages exist for edge cases (headless servers, non-Anthropic SDKs, custom tool wrappers).

Notes

  • The MCP server only runs while the axiom.ai desktop app is open. Quitting the app stops the server.
  • Each API call from the LLM counts against the standard rate limits and runtime allowance. LLMs that poll aggressively can trip rate limits; if you see 429 errors in the desktop app logs, tell the LLM to slow down.
  • The MCP server is local-only. It listens on a port on your machine and isn't reachable from the public internet.
  • API keys are stored in the desktop app's encrypted local config, not in plaintext on disk.
Next
Build your own server (Python)
On this page