MCP Overview

What the CueAPI MCP server is and why it exists.

What this is

@cueapi/mcp is the official Model Context Protocol server for CueAPI. It exposes the five coordination primitives (Cue, Execution, Outcome) as tool calls inside any MCP host, so an agent can schedule its own follow-up work and report outcomes with evidence from inside a conversation.

Once configured, the agent gets eight tools — create, list, get, pause, resume, delete cues, list executions, and report outcome — alongside whatever other tools the host provides.

Why

The gap MCP closes is the one between "the model decides it should run again in three hours" and "something in the world actually makes that happen." Without a scheduler the agent can only ask the user to re-prompt. With CueAPI behind MCP, the agent creates a cue, the cue fires on time, a handler runs, and the agent reports the outcome back to the same CueAPI account — evidence attached — so every scheduled run is queryable as an accountability record.

Transport

Today: stdio only. The host spawns @cueapi/mcp as a subprocess and talks to it over stdin/stdout. That means the server runs on whatever machine runs the host — Claude Desktop on your laptop, Claude Code in your terminal, Cursor in your IDE.

HTTP transport (remote MCP for Claude.ai web, ChatGPT, etc.) is on the roadmap. See Claude.ai web for the current workaround.

Tools exposed

Eight tools, one per CueAPI surface the agent actually needs. Full schemas and examples in Tools.

ToolPurpose
cueapi_create_cueCreate a cron or one-time cue
cueapi_list_cuesList cues, filter by status
cueapi_get_cueFetch one cue by ID
cueapi_pause_cuePause a cue
cueapi_resume_cueResume a paused cue
cueapi_delete_cueDelete permanently
cueapi_list_executionsHistorical execution records
cueapi_report_outcomeReport outcome with evidence

Install

Pick your host:

Source

How do I know if my agent ran successfully?
Ctrl+K