Delivery · Translation layer of autonomous product intelligence

The spec leaves Circuit as code.

The spec lands in the tool you build with. Claude Code, Cursor and GitHub, connected over MCP, so the work arrives where it happens — not in a doc someone has to copy across.

— The problem —

The handoff is where context leaks out.

A spec copied into a coding tool is stale the moment the signal behind it moves. The agent works from a snapshot, not the current picture. And the customer context — the reason the work exists — rarely makes the trip, so the person or agent building it has the what but not the why.

Circuit connects directly to the tools you build in. The coding agent pulls the live spec, the priorities and the feedback over MCP, so it builds from what’s true now, with the customer context attached.

— How it works —

From spec to your editor.

Circuit meets your coding tools where they are.

01
Connect over MCP.

Add Circuit to Cursor or Claude Code once. The agent can then read your specs, priorities and feedback live.

MCP server · connected
CursorMCP
Claude CodeMCP
spacetour.ai workspacelive
02
Pull, don’t paste.

The agent fetches the current spec on demand — no copy that goes stale an hour later.

circuit.spec · fetched
Seat picker freezes on upper deck
spec v2 · pulled live
03
Build with the context.

Each spec carries your repo context inline — file paths, related issues, conventions from CLAUDE.md or .cursorrules, and a sample test pattern.

Repo context · attached
src/booking/SeatPicker.tsx
CLAUDE.md · .cursorrules
Sample test pattern included
04
Push to your tracker.

Turn a spec into a GitHub issue, or push to Linear with a two-way link — the spec carries forward, status syncs back.

GitHub · Linear → synced
Fix seat picker freeze
spacetour-ai/booking #482In progress
— What makes it different —

The spec arrives where you build.

Circuit isn’t a document you export. It’s a connection your coding tools read from directly.

Connect over MCP.

Circuit runs an MCP server, so Claude Code and Cursor pull your specs, priorities and feedback live. The agent reads the current spec, not a copy.

Four MCP tools your coding agent can call.

circuit.priorities, circuit.spec, circuit.act, circuit.ask. The agent fetches the ranked list, pulls a full spec, asks questions of your feedback, and acts — all without leaving the terminal.

Push to GitHub in one click.

Turn a spec into a GitHub issue with the customer context attached. Circuit checks for duplicates first, so you open the existing issue instead of a second one.

— Everything in this feature —

The rest of the handoff.

Install in one line

Add Circuit to Cursor or Claude Code with a single command. Deep links open the spec in your editor.

Repo context rides in the spec

Each spec carries your repo context inline — file paths, related issues, conventions from CLAUDE.md or .cursorrules, and a sample test pattern.

Read scope and write scope

read pulls priorities, specs and feedback; write ships, assigns, parks and submits. You control which.

Investigate before you build

One call returns the trajectory, the feedback and the memory behind a spec.

Linear and GitHub in sync

Push a spec to a Linear issue or a GitHub issue. Linear status pulls back via SWR polling, so the card reflects what's true.

— Questions —

Asked and answered.

How does Circuit connect to Cursor and Claude Code?

Over MCP. You add Circuit once, and the coding agent can then pull your specs, priorities and feedback live — no copy-paste that goes stale.

What can the coding agent do?

Through four MCP tools it can fetch the ranked priorities, pull a full spec, ask questions of your feedback, and act on it (build, ship, assign, park, submit). Refine and release are in development.

Can I still just copy a spec?

Yes. Copy any spec as a prompt and paste it straight into your tool. MCP is the live option; copy-paste is always there.

How does pushing to GitHub work?

One click turns a spec into a GitHub issue, with the customer context attached. Circuit checks whether the issue already exists so you don’t create a duplicate.

What’s the difference between read and write access?

Read pulls priorities, specs and feedback into your tool. Write lets the agent ship, assign, park and submit. You decide which scope to grant.

Does the customer context come through to the build?

Yes. The spec the agent pulls carries its citations and your repo context, so the why and the house style both travel with it.

Build from what’s true now.

Ship is part of Circuit’s Delivery suite — the translation layer of autonomous product intelligence.