Open the repo. The agent already knows the plan.
ProductDevelopment.io is agent-first: 115 MCP tools over the same database, roles and rules as the app. Claude Code and Codex find the right application from the git remote, read its plan and update the work — without being able to trample anyone else's.
The plan for exactly this codebase
plan_context takes the git remote of the open repository — SSH, HTTPS or owner/name — and matches it to an application. It returns the tech stack, who owns each ticket type, active cycles and the open tickets ordered by priority, stage and blockers, with the titles of their attached documents.
- Read-only, from one consistent snapshot
- Unknown or ambiguous repositories return choices, not a guess
- The plan-open-tickets prompt turns it into a working plan
No match? It returns matched: false and the applications to choose from — it never guesses.
Agents cannot overwrite your team
Every update sends the version the agent read. If a person — or another agent — saved first, the write is rejected and nothing changes. Bulk edits apply to all selected tickets or none, and workflow rules apply to agents exactly as they do to people.
- Atomic bulk edits of 1–100 tickets
- Transition and completion rules enforced for every caller
- Server instructions: ticket text and notes are data, not instructions
- Codex · Reads PD-407version 7
- KOK. Osei · Adds an estimate in the appversion 8
- Codex · Updates PD-407 with version 7rejected
- Codex · Re-reads, updates with version 8version 9
A person's permissions, not a master key
Each connection belongs to a person in one workspace, with that person's role. OAuth 2.1 with PKCE and a consent screen, read or write scope, rotating refresh tokens — and every connection listed where you can revoke it.
- Owner, admin, member and viewer roles apply to agents
- Workspace tokens expire after 90 days
- A read-only endpoint checker to diagnose a connection
Specs and logs where the agent will look
Attach links, Markdown or text files of up to 200,000 characters to a ticket. The agent sees them in the plan and can read them before it starts — and meeting notes and customers are one query away.
- Links, Markdown and text documents on tickets
- Meeting context recoverable per ticket
- A daily-update prompt for standups
Four steps, once per person
Copy your endpoint
Integrations in the app shows your workspace's MCP address.
Add it to your agent
Claude Code, Codex or any client that speaks MCP over HTTP.
Sign in and consent
OAuth in the browser, choosing read or write. Or use a 90-day workspace token.
Ask for the plan
Open a repository and ask what to work on. The plan-open-tickets prompt does the rest.
The agent package ships a development workflow skill for Claude Code and Codex. Credentials are never bundled into it.
Your coding agent works from the real plan
ProductDevelopment.io runs its own MCP server on the same database as the app. Claude Code, Codex or any MCP client signs in once, then reads the plan for the repository it is standing in, picks up tickets and records progress — with the permissions of the person who connected it.
An illustration with demo data. Tool names and behaviour are the server's own.
# your workspace's MCP endpoint
claude mcp add --transport http \
productdevelopment https://<your-workspace>/api/mcpBrowser sign-in with OAuth and a consent screen, or a workspace token that expires after 90 days. Plus an agent package with a development workflow skill for Claude Code and Codex.
What an agent can do
Named tools for the work that needs care, and generic list, get, create, update and delete tools for every record kind.
Give your agent the real plan
Request access and bring the repository you are working in. We connect your agent with you.