Where AI specs go
Cursor and Claude Code via MCP
Circuit is the feedback MCP for Cursor and Claude Code — the feedback to spec pipeline, delivered via the Model Context Protocol. Pull priorities, fetch specs, search feedback and take actions from inside your coding tool. Use circuit.act to mark as shipped, notify customers, submit feedback or submit a transcript. Ship from terminal.
Setup
- •Cursor: Add Circuit MCP to your Cursor config
- •Claude Code: Run
claude mcp add circuit - •First run opens a browser for OAuth authentication
- •Token cached locally for 30 days
4 MCP tools
circuit.prioritiesRanked priorities ready to buildcircuit.specFull spec for any priority with all 5 sectionscircuit.askSearch feedback and priorities for contextcircuit.actShip from terminal: mark shipped, notify customers, submit feedback or transcripts, assign, correct, parkExample: asking for priorities in Cursor
GitHub. Codebase comprehension for every spec
Connect your repo and Circuit reads the codebase to generate specs that reflect how your team builds — not just file paths, but what's inside them. Specs can only reference files that exist in your codebase. The system won't generate paths it hasn't seen. Circuit reads your .cursorrules and CLAUDE.md before generating, so the spec your coding tool receives already follows your team's rules. Create GitHub issues directly from any spec. The result is a codebase-aware product spec — not a generic template.
What Circuit reads
- File signatures — exports, types and key functions from relevant source files
- .cursorrules / CLAUDE.md — specs follow your team's coding rules
- Testing patterns — a sample test file so specs match your testing style
- Directory structure — compact file tree for accurate path suggestions
- Open issues — avoids duplicating existing tracked work
- Recent merged PRs — informs which files are suggested as relevant
- Tech stack — detected framework, language and environment
- CI/CD + config files — workflow names and config files present
- README — project overview when no AI config files exist
All read-only. Circuit never writes to your repository.
What specs include
Actual file paths marked as new or modify, coding conventions that follow your .cursorrules and CLAUDE.md rules, and a Heads Up section when similar open issues already exist in your repo.
As your team refines .cursorrules and CLAUDE.md over time, Circuit's specs improve automatically — both systems read from the same source of truth.
Where feedback comes from
Slack. Automatic feedback collection
Connect your workspace and Circuit monitors selected channels with no manual forwarding. OAuth connection with AI-suggested channel selection. Polled every 10 minutes. Messages parsed for customer metadata and revenue band.
Slack setup guide →Google Sheets. Bulk import
Connect Google Drive, select a spreadsheet, review the auto-detected column mapping and import. Useful for historical feedback or tools that export to Sheets.
Google Drive setup →CSV. Drag and drop
Upload any CSV. Circuit auto-detects delimiter, encoding and headers. Map columns, review and import.
Surfaces. Five ways to collect feedback on your site
One script tag, five surface types. No authentication needed for submitters. Each surface is a 1-line embed with a live preview editor and per-surface analytics.
Surface types
BubbleMost responses. Best for open-ended feedback.EmbedSubtle. Works well in docs and help pages.BannerHigh visibility. Good for announcements or surveys.ThumbsThumbs up/down with optional written follow-up. Per-feature tracking with data-context.TriggerZero footprint. Best in forms or settings.
Use data-context to tag feedback by page, flow or feature — filter it in Surface Analytics later. Call .identify() from your JS to pass customer email and revenue tier automatically.
Transcripts. Customer calls into priorities
Upload call recordings or interview transcripts and Circuit extracts feedback segments automatically — up to 25 per call, with speaker attribution. Supports Otter.ai, Fireflies, Grain and Whisper exports. Import directly from Google Drive. A research interview feeds the same priority pool as widget feedback.
Manual entry
Paste feedback from anywhere. Email threads, support tickets, meeting notes. Processed through the same AI pipeline as every other source.
API. Submit feedback programmatically
Single and batch endpoints for submitting feedback from your own backend, support tools or data pipelines. Supports deduplication, customer metadata and revenue tier. See API docs →
Automation and events
Webhooks. Connect anything else
Send events to any HTTP endpoint, Slack channel, Linear, GitHub or PagerDuty.
Available events
- •Spec generated
- •Build status changed (ready → building → shipped)
- •New priority created
- •Priority volume spike (>20% increase)
- •New feedback received
Test any hook before going live with full execution history, timestamps and status codes. All integrations included on every plan.
See pricing →