Claude.ai Web and ChatGPT

Where MCP runs today

The CueAPI MCP server communicates over stdio — stdin and stdout with the host process. That requires a host that can spawn local subprocesses:

• Claude Desktop (macOS, Windows, Linux)
• Claude Code (terminal)
• Cursor, Zed, and other desktop MCP hosts

Browser-based agents like Claude.ai and ChatGPT cannot spawn subprocesses on your machine, so the stdio transport does not work there.

What's on the roadmap

Remote MCP over HTTP/SSE is an MCP spec feature and the ecosystem is moving toward it. When the CueAPI MCP server ships an HTTP transport, web-based agents will be able to connect to a hosted endpoint (e.g. https://mcp.cueapi.ai) and use the same eight tools.

Until then, web agents can still use CueAPI — just through the REST API, not through MCP.

Using CueAPI from a web agent today

Two patterns work well.

Pattern 1: agent calls the REST API directly

If your web agent supports custom tool / API calling, point it at CueAPI's REST surface:

POST https://cueapi.ai/v1/cues
Authorization: Bearer cue_sk_your_key_here
Content-Type: application/json

{"name": "daily-digest", "cron": "0 9 * * *", "callback_url": "https://..."}

The eight MCP tools are a curated subset of the REST API — anything MCP can do, the REST API can do too. See the API Reference.

Pattern 2: use a CueAPI-backed agent on desktop, let it schedule work that runs anywhere

Many teams use Claude Desktop or Claude Code locally as the "planner" — the agent that decides what to schedule and reports outcomes — while the actual cue handlers run on a worker daemon, a serverless webhook, or another agent entirely. The planner talks MCP; the handlers talk REST. Both touch the same CueAPI account.

Following the roadmap

HTTP transport work is tracked on the cueapi-mcp GitHub repo. Subscribe to releases or watch the repo.