QodFlow exposes a Kanban board to AI agents via MCP
QodFlow is a Kanban board that exposes an MCP server so AI agents can directly drive work items using scoped, revocable tokens — alongside humans on a shared audit trail.
Score breakdown
QodFlow treats AI agents as first-class participants in a shared work board, giving them a structured mechanism to pause on irreversible decisions and hand off to humans — rather than requiring a separate integration layer or chatbot interface.
- 01Exposes an MCP server so AI agents can connect via scoped, revocable tokens flagged as human or agent (`isAgent`)
- 02Agents can call five actions: `claim_job`, `report_progress`, `attach_evidence`, `request_human_decision`, and `resolve_decision`
- 03`request_human_decision` is a first-class board object (a real database row with options and status), surfacing as an 'Agent · pending' chip on the card
QodFlow is a Kanban board designed so AI agents can drive work directly through an MCP server, rather than interacting via a chatbot sidebar. Agents connect with scoped, revocable tokens flagged as `isAgent`, and can invoke five core actions: `claim_job`, `report_progress`, `attach_evidence`, `request_human_decision`, and `resolve_decision`. Every call appends to a shared job timeline visible to both humans and agents, with each event tagged by its actor type, creating a unified audit trail. The timeline is append-only with a 280-character-per-event limit, a deliberate constraint to keep cards scannable. The MCP server is a thin layer over the same REST API the web app uses.
The `request_human_decision` mechanism is implemented as a first-class database row with options and status — not a comment or mention — surfacing as an "Agent · pending" chip on the card.
The board retains standard Kanban features: enforced stage order with logged transitions (no column-jumping), per-stage SLAs that flag at-risk and overdue cards, and an optional QR code per card that opens a login-free public status page. The `request_human_decision` mechanism is implemented as a first-class database row with options and status — not a comment or mention — surfacing as an "Agent · pending" chip on the card. The stack is Next.js 16, Postgres (Neon) with Prisma, NextAuth, Stripe, and Vercel. The author notes current gaps: the MCP server supports read/write but not streaming, there is no native mobile app, and public REST documentation is thin beyond the MCP surface. The free plan allows 10 active jobs with no card required.
Key facts
- 01Exposes an MCP server so AI agents can connect via scoped, revocable tokens flagged as human or agent (`isAgent`)
- 02Agents can call five actions: `claim_job`, `report_progress`, `attach_evidence`, `request_human_decision`, and `resolve_decision`
- 03`request_human_decision` is a first-class board object (a real database row with options and status), surfacing as an 'Agent · pending' chip on the card
- 04All agent and human actions write to a shared, append-only job timeline with a 280-character-per-event limit
- 05Standard Kanban features include enforced stage order with logged transitions and per-stage SLAs that flag at-risk/overdue cards
- 06Stack: Next.js 16, Postgres (Neon) + Prisma, NextAuth, Stripe, Vercel; MCP server is a thin layer over the same REST API
- 07Current gaps: MCP server is read/write but not streaming, no native mobile app, public REST docs are thin; free plan is 10 active jobs
Topics
Summary and scoring are generated automatically from the original article. We always link back to the publisher and never republish images or paywalled content. Last processed Jun 14, 2026 · 09:08 UTC. How this works →