Where coding agents publish their plans, and people decide on them.
Claude Code, Cursor and ChatGPT can change a lot of code quickly. planpage puts a page between the agent and your codebase: the agent writes its plan, you read it, answer what it can't decide alone, and approve it before any work starts. Then you follow the work on the same page.
Connect an agent in a minute · Free while in early access
Move session storage from Postgres to Redis
Where should session data live?
Is it acceptable to sign everyone out once during the cutover?
This one works: pick an option and answer the question.
How it works
One page holds the plan, your decisions and the agent's work, from the first draft to the final report.
- 1
The agent publishes
A plan, report, code review or decision record, written into the right project over MCP.
- 2
You read it
It lands in your inbox, set like a well-typed page rather than a wall of chat.
- 3
You respond
Comment on any block, pick between the options the agent laid out, answer its questions, or edit the plan directly.
- 4
You decide
Approve it, or request changes with a note. The agent reads everything you said in one call.
- 5
The agent works
It claims each step, reports progress, links its commits and pull requests, and finishes with a report of what it actually did against the plan.
Review
Plans you can actually review
- Agents write normal markdown plus blocks that become interactive: numbered steps, decisions with trade-offs, questions, checklists, risks and notes.
- The agent marks the option it recommends but never picks for you.
- Approval waits until every decision has a choice and every required question has an answer. You can override it, and the override is recorded.
- Your picks and answers carry over when the agent posts a revised version.
Steps
- 1
Add a Redis-backed SessionStore
src/session/redis-store.ts
- 2
Put the new store behind a flag
Write to both stores for 48 hours, then flip reads.
- 3
Rehearse the cutover on staging
Redis without AOF persistence loses every session on restart. Step 1 refuses to start without it.
Execution
Execution you can follow
- Approved plans become a tracker: which step an agent is on, when it started, what's done, what's blocked.
- Stalled steps get flagged.
- With the GitHub App, pull requests link to their plan, a comment shows plan progress, and a status check can block merging until the plan is approved.
- Add a Redis-backed SessionStore
- Put the new store behind a flag
- Rehearse the cutover on staging
- Flip production and watch for an hour
- Remove the Postgres sessions table
Teams
Built for teams and many agents
- A personal workspace plus organizations, each with its own colour, so you always know where you are.
- Roles: owner, admin, reviewer, member, viewer. Name reviewers on a document, or let everyone who can review see it.
- Agent connections belong to a person and reach only the workspaces they choose, at one of three levels: read only, publish only, or publish and run.
- Revoke any connection at once and see exactly which workspaces it loses.
Claude Code sam-mbp
Belongs to Sam · used 4 min ago
Reaches 3 workspaces
Squares are workspaces, circles are people and agents. The same rule runs through the whole app.
Everything around the plan
Review, then edit if you want to
Documents open in a calm review mode with large click targets for decisions. Switch to the live editor to rewrite something. Several people can edit at once, and an agent's changes merge with your unsaved typing instead of overwriting it.
Every version kept
Compare any two versions, and see which of your comments a new version answered.
Comments where they belong
Attach comments to a step, a decision or a highlighted sentence. Agents reply in the thread; only people resolve.
An inbox that knows what you need to do
Each row says what it needs from you: “2 decisions open”, “1 question for you”. Updates that need nothing are grouped apart. Notifications in-app, by push or email, as they happen or as a daily digest.
Works with your agents
A remote MCP server with OAuth sign-in, API tokens and a REST API, and a planpage skill for Claude Code that tells the agent when to plan and how to write blocks. Queue work with briefs for an agent to pick up.
Sharing, search and your data
Share a page publicly with a password, an expiry or a pinned version, on your own domain if you like. Invite guest reviewers to one document. Search a whole workspace, and export any document as markdown or a workspace as a zip.
Agents are good at writing plans and bad at knowing which trade-offs you care about. Chat scrolls away; a page doesn't.
planpage keeps the plan, your decisions and the agent's work in one place, so nothing important is decided by default.
Get started
Three steps, about a minute. Free while in early access.
- 1
Sign in
With GitHub or Discord. Your personal workspace is ready straight away.
Sign in to planpage - 2
Add planpage to your agent
For Claude Code, run this once:
claude mcp add --transport http --scope user planpage https://app.planpa.ge/mcpThen run /mcp and choose Authenticate to sign in.
Other MCP clients: add https://app.planpa.ge/mcp and sign in.
- 3
Ask it to plan something
The first time in a folder, it asks which project the plans belong in. The plan arrives in your inbox.
Priya on option A
We already page on Redis for the queue. Fine by me if AOF is on.