AI Agent

Requence ships a dedicated agent MCP server that lets you connect any MCP-capable AI client (Claude Code, Claude Desktop, Cursor, and others) to your Requence instance and work with it conversationally. It is separate from the execution MCP server, which exposes your task templates as callable tools — the agent server is for talking about and working on your instance instead.

Point your client at one URL, authenticate with your own personal access token, and the agent orients itself from there — it needs no extra setup or prompting.

Page	What it covers
Installation	The server URL, the token it needs, and the config for Claude Code, Claude Desktop, and Cursor
Investigating Runs	Read-only: why a run failed or is stuck, what it returned, and where its time went
Building Workflows	Guardrailed writes: creating and editing templates, services, and logic scripts — live on your canvas or headlessly
Permissions & Safety	What the agent can and cannot do: your token, the audit marker, and why it never saves on live
Skills	Two optional, purely advisory skills that package good call sequences

The two halves

Reading and writing are worth keeping apart in your head, because they carry different risk.

Investigating a run only reads. Ask “why did the nightly sync fail?” or “where did the time go?” and the agent walks the execution tree — sub-tasks included — to the node that actually caused the problem, and answers with deep links into the UI for every hop.

Building a workflow writes, but never the final say. The agent joins the same collaborative editing session your browser uses, so you watch it work on your canvas; it type-checks itself before committing; and it never saves on live and never merges. See Permissions & Safety for the whole boundary.