For teams that write code with AI agents

A product plan your AI agent reads

Describe your idea in a chat — AppKeel turns it into a vision, a roadmap and tasks. Claude Code or another agent picks a task up right from here, and you accept the result from its report without reading the code.

Start free See how it works

Needs your own AI key or a Claude subscription — connected in a minute after you sign in.

Works withClaude CodeCursorAny MCP client

canon · payments-service

Roadmap

Vision

Tasks

Decisions

Data model

Scenarios

PHASE 02

Sync engine

✓Define canon schema

✓Expose plan over MCP

Diff proposed changes to review

AGENT · CLAUDE CODE

Proposed 3 changes against the plan

Review diffLater

The canon

One plan the whole build stays true to.

Stop scattering the spec across notes, tickets and chat threads. Appkeel holds the single source of truth — and keeps your agents pointed at it.

Living canon

Vision, roadmap phases, tasks, locked decisions, data model and user scenarios — structured, versioned, in one place.

Read over MCP

Agents connect once and read the current plan on demand — always building against the truth, never a stale copy.

Human review

Agents never edit the canon directly. Every proposed change comes back as a revision for approval — and if an agent hits a gray area, it raises a question in the task thread instead of guessing.

Plan ≠ code

Planning lives apart from the repo but stays fully in sync — restructure the roadmap without touching a line of code.

Decision log

Architectural calls are recorded as they're made, so agents inherit the why — not just the what. Locked decisions always win over aspirational vision or stack prose — agents build to the decision, not the pitch.

Designs

Schematic screen prototypes anchored to your data model and user workflows — so agents build UI against an approved layout, not a guess.

How it works

Plan once. Ship in a loop.

00

Start from an empty folder

Point your agent at an empty folder — AppKeel gives it a CLAUDE.md, a set of skills and setup steps before the first line of code.

01

Write the canon

Capture the product vision, roadmap phases and data model. Appkeel turns it into a structured plan agents can query.

02

Your agent builds against it

Claude Code and friends connect to Appkeel over MCP, claim a ready task and implement it in your own repo. The agent is yours: the product holds the plan and takes the result back — you run the agent.

03

You review the diff

The agent submits the task with a two-axis review report; proposed changes to the plan wait in the Inbox. Accept what is right, send back what is not — without reading the code.

Connect in one line

Point your agent at the keel.

Add Appkeel as an MCP server and your agent can read the plan, phases and decisions — and open changes for review — no glue code.

✓ Read-only by default, scoped tokens

✓ Every write becomes a reviewable diff

✓ Works with any MCP-compatible client

✓ Multi-project — export any project as a portable JSON bundle

.mcp.json

{
  "mcpServers": {
    "appkeel": {
      "command": "npx",
      "args": ["-y", "@appkeel/mcp"],
      "env": { "APPKEEL_TOKEN": "ak_live_…" }
    }
  }
}

Give your agents something solid to build on.

Set up your first canon free. Connect an agent in minutes.

Start free See the MCP setup