Pocket PMOGive your AI agent real project maths, not a hallucinated Gantt chart.
Pocket PMO exposes PERT, CPM, Monte Carlo, EVM and risk scoring over an MCP server, so AI planning agents compute answers instead of improvising them.
The published market stocks records. Nobody stocks computation.
Project-management MCP servers on the market today connect assistants to tasks, boards, pages and issues. That lets an agent create a plan, and then describe it. What we could not find anywhere in that market is a server that tells the agent how likely the plan is to succeed. PocketPMO’s MCP server is built for exactly that missing half: computation, not storage. Existing servers create and update tasks, boards, issues and pages, read project state back, summarise data in prose, and estimate dates by language intuition. None of them compute. Pocket PMO's server carries the computational core: PERT three-point estimating with risk-adjusted outputs, Monte Carlo delivery probability for a named date, CPM critical-path identification, and EVM, risk scoring and portfolio roll-ups.
An agent asks; the engine answers.
Shape of a session (illustrative request and response; exact tool names are published with the public docs; endpoint and authentication details are not yet published). Eight read-only tools: PERT estimates, ROM estimates, Monte Carlo simulation, CPM critical path, EVM, risk scoring, single-project and PMO portfolio queries.
The technical buyer’s reasons to care
Your agent produces plans. Give it a probability engine so its answers carry numbers a human can check, and stop it improvising confidence levels in prose. Ask your assistant “if we add this scope, what happens to the date?” and get a delta computed by the same engine the human-facing app uses. Check the market: records everywhere, computation nowhere. If your agents must reason about delivery risk, this is the only server of its kind today. Preview status, stated plainly: the capability list is final, and the computation core is the same one running the no-signup sizer on the home page. Public endpoint documentation, access details and directory listings (mcp.so, Glama, Smithery, PulseMCP and the awesome-mcp-servers project/task-management list) follow.
Run the example on your own schedule
Try the computation core free in Pocket PMO: the same engines the MCP server exposes — PERT, CPM, Monte Carlo — running in your browser.