Microsoft Project Online retires September 30, 2026, migrate to a modern platform before it's too late.Start migration
Onplana MCP server, live at mcp.onplana.com

Project management that lives inside Claude, ChatGPT, and Cursor.

Onplana ships a public Model Context Protocol server with 29 curated tools, hybrid BM25 + vector search across your org's indexed content, OAuth 2.1 or PAT auth, plan-gated access, and end-to-end audit. Built on the open MCP standard from Anthropic. Works with Claude Desktop, ChatGPT, Cursor, Gemini CLI, Gemini Code Assist, GitHub Copilot, Grok, and any other MCP client that speaks Streamable HTTP. Two-minute setup. Free plan.

29 tools, plan-gatedOAuth 2.1 or PATAudited end to endFree plan

Same prompt. Different client. Same underlying tool call.

The mockups below show the shape of the response when you ask the same question, "What's overdue in Onplana?", in Claude Desktop and in ChatGPT. The tool-call indicator at the top of each response is the visible proof that the AI is reading live data from the MCP server, not paraphrasing. Same data on both sides, different presentation idioms per client.

Claude Desktop
What's overdue in Onplana?
Searched available tools

Pulled the overdue list from your Onplana org. Eight tasks open, all in the same project, all assigned to you:

Overdue, Customer onboarding revamp, 8 tasks:

TaskDaysStatus
Q3 product roadmap review5In progress
Customer success playbook v25In progress
Vendor security audit5To do
Pricing page redesign4To do
Quarterly OKR check-in3To do
Onboarding email cadence v32In progress
+ 2 more, due today
ChatGPT, custom connector
What's overdue in Onplana?
Called tool: list_overdue_tasks

Here are the overdue tasks in Onplana:

  1. Q3 product roadmap reviewIn Progress
    Project: "Customer onboarding revamp" · Due 5 days ago
  2. Customer success playbook v2In Progress
    Project: "Customer onboarding revamp" · Due 5 days ago
  3. Vendor security auditTo Do
    Project: "Customer onboarding revamp" · Due 5 days ago
  4. Pricing page redesignTo Do
    Due 4 days ago
  5. + 4 more, including 2 due today

Mockup using anonymized task data. The chrome (tool-call indicator, conversation layout, response idioms per client) matches what you see when you wire up the real MCP server. Run the integration on your own org to see your own data in the same shape.

What is MCP project management

MCP project management is project management work done inside an AI assistant rather than inside a PM tool's UI. Model Context Protocol (MCP) is the open standard from Anthropic that lets AI clients read and write data in real systems through a typed tool catalog. The AI doesn't guess. It asks the tool, gets structured data, and answers from facts.

The shape of the workflow is: you stay in the AI client you already use, Onplana sits behind it as a tool source. You ask "what's on my plate this week" in Claude, Claude calls the relevant Onplana tools, returns the answer. You say "mark Task A as done" in ChatGPT, ChatGPT calls the update tool, the change lands in your real org. There is no copy-paste between the AI and the PM tool, and there is no synthetic "demo data" — the AI reads and writes your live workspace.

Why this matters now: agentic AI clients are the fastest-growing surface for knowledge work. When the AI sits between you and your tools, the tools that have native MCP support get used and the tools that don't get abandoned for the ones that do. Onplana ships native MCP today, with 29 curated tools, hybrid search across indexed content, and audit-grade traceability on every call. The two-minute setup below gets you from a fresh account to "what's overdue?" answered by Claude in less time than it takes to read this paragraph.

Two-minute setup

Four steps. The slowest step is signing up for an Onplana account if you don't already have one — the rest is config paste-and-restart. Two clients shown side-by-side; the install pages for every other supported client live under /mcp.

1

Mint a personal access token

In Onplana: Settings → Developer → Personal Access Tokens → Create. Scope: MCP_AGENT. Name it after the client you're installing into (e.g. "Claude Desktop laptop") so you can revoke per-device later. Copy the token — it shows once.

2

Paste the server config

Claude Desktop

Settings → Developer → Edit Config. Add:

{
  "mcpServers": {
    "onplana": {
      "url": "https://mcp.onplana.com/mcp",
      "headers": {
        "Authorization": "Bearer <PAT>"
      }
    }
  }
}

ChatGPT custom connector

Connector picker → Add custom. Use the OAuth flow (preferred) or paste the PAT under "Bearer token" auth:

Server URL:  https://mcp.onplana.com/mcp
Transport:   Streamable HTTP
Auth:        OAuth 2.1 (recommended)
             OR Bearer + PAT
3

Restart the client

Claude Desktop and Cursor pick up the config on launch. ChatGPT custom connectors are live the moment the OAuth handshake completes — no restart needed.

4

Test with "What's overdue?"

Open a new chat. Ask "What's overdue in Onplana?" The client shows a tool-call indicator (Claude says Loaded tools, ChatGPT saysCalled tool), then returns your actual overdue task list. The proof block above is what the response looks like in each client. If you see a generic "I don't have access" answer, the auth isn't wired, check the PAT scope.

Other clients? See the install docs at /mcp, the dedicated Claude and Gemini how-to pages, or the engineering deep-dive at /mcp/how-we-built-it.

10 prompts that work

Copy any of these into Claude, ChatGPT, or Cursor with the MCP server connected. Each prompt maps to a specific Onplana tool call; the AI handles the decomposition. Direct + specific + action-oriented prompts win.

Any client
"What's overdue in Onplana?"

Maps to: list_overdue_tasks

Any client
"Show me sprints ending this week and what's still open in each."

Maps to: list_sprints + list_tasks

Any client
"Create a task in the Onboarding project: "Draft welcome email sequence", due Friday, assign to me."

Maps to: create_task

Any client
"What risks are flagged HIGH or CRITICAL across my projects?"

Maps to: list_risks

Claude
"Draft a status report for the Onboarding project covering last week's progress, blockers, and next-week plan."

Maps to: list_tasks + activity_summary

ChatGPT
"Summarize what changed in the Onboarding project since I last checked in (three days ago)."

Maps to: list_activity

Any client
"Mark task "Draft welcome email sequence" as done."

Maps to: update_task_status

Claude
"Show the critical path for the Q3 Launch project."

Maps to: get_project + list_tasks

Any client
"Who has more than 30 hours of work assigned next week?"

Maps to: capacity_report

Any client
"Search Onplana wiki for "migration runbook" and summarize the latest version."

Maps to: search_content (hybrid BM25 + vector)

Five real workflows

The shape of daily PM work after MCP lands. None of these are hypothetical, they are the patterns Onplana team members use against their own production org.

Daily stand-up via AI

The intent: A 30-second status check before the team meeting. Instead of clicking through five project boards, ask one question.

The example: "What's overdue across my projects, and what's due today?" returns a single tabulated list with project context and assignees. Paste straight into the stand-up doc, no copy-paste between tools.

Maps to: list_overdue_tasks + list_tasks_due_today

Sprint planning in Cursor (while reading the code)

The intent: For engineering teams. The PM context lives a tab away from the code; switching breaks flow.

The example: "Add the OAuth retry logic feature as a task in the Backend Q3 sprint, estimate 8 hours, depends on 'PKCE storage hardening'." Cursor calls the tool, the task lands in the sprint with the dependency wired, you go back to reading the diff.

Maps to: create_task + link_dependency

Weekly status report draft

The intent: The Friday afternoon write-up that nobody enjoys. The AI has full read access to the project's activity history.

The example: "Draft a status report for the Customer Onboarding project: what shipped, what slipped, RAG status, next-week plan." Claude/ChatGPT pulls the activity log, last-week completions, in-progress tasks, and blockers; returns a clean RAG-formatted draft. Edit-in-place rather than write-from-scratch.

Maps to: list_activity + list_tasks + list_risks

Risk triage before the steering committee

The intent: Surfacing the items that need executive attention vs. the items the project team can handle.

The example: "List all HIGH and CRITICAL risks across the portfolio, grouped by impact area. For each, summarize the latest update and what's blocking resolution." Returns the right level of detail for a 20-minute exec review.

Maps to: list_risks + list_risk_updates

New-team-member onboarding

The intent: A new hire asks "what does our team work on?" — a question that takes longer to answer well than it sounds.

The example: "Summarize the active projects in my Onplana org, who's on each, and what shipped this quarter." Returns a digestible team-orientation primer. Send to the new hire on day one as their reading list.

Maps to: list_projects + list_team_members + list_activity

Multi-client compatibility

One MCP server, eight verified clients. Switching clients is a config-file change; the tool catalog, scopes, and behavior stay identical. No per-client integration code on our side.

ClientInstallOAuth 2.1PATNotes
Claude DesktopJSON snippet (Developer → Edit Config) + restartOAuth via Connectors Directory + JSON for desktop
Claude.ai (web)One-click via Connectors Directory·OAuth only; PAT path is desktop-specific
ChatGPT custom connectorConnector picker → "Add custom" → OAuth flowAvailable on Plus / Pro / Team / Enterprise / Edu
ChatGPT Codex CLIAdd server URL + token to .codex-cli config·CLI-only; PAT recommended
Cursor~/.cursor/mcp.json snippet + restart·Same JSON shape as Claude Desktop
Gemini CLIgemini extensions install (one line)Easiest install path of any client
Gemini Code AssistVS Code or JetBrains extension settingsInside the IDE, no terminal needed
GitHub CopilotVS Code settings → MCP servers·Preview feature; check your org's rollout
GrokCustom MCP server config in Grok settingsMost recent client; minimal install drift

Any other client that speaks the MCP Streamable HTTP transport works against the same endpoint with no changes on Onplana's side. See the open MCP standard from Anthropic for protocol details.

Without MCP vs. with MCP

The shape of the same daily-PM tasks before and after the MCP server is wired in. The unit of comparison is "minutes per day spent context-switching between the AI client and the PM tool."

Without MCP

  • Switch tabs to Onplana to find overdue tasks
  • Copy-paste the list back into Claude for triage
  • Ask Claude to draft a status report from the pasted data
  • Switch tabs to update each task's status manually
  • Repeat for every project, every day
  • ~20-30 minutes per day of context-switching tax

With MCP

  • Ask Claude or ChatGPT directly — it reads your real org data
  • Status report drafted from live activity, not pasted text
  • Mutations land via the same conversation, preview-then-confirm
  • Undo via the AiOperation ledger if anything goes sideways
  • Audited end to end, scoped per token
  • ~30 seconds per day for the same surface area

Frequently asked questions

Ten questions PMOs ask most often when evaluating MCP-based AI workflows. Embedded as FAQPage schema so they surface in Google's "People Also Ask".

What is MCP project management?

Project management work done inside an AI assistant rather than inside a PM tool's UI. Model Context Protocol (MCP) is the open standard from Anthropic that lets AI clients like Claude Desktop, ChatGPT, Cursor, Gemini, and GitHub Copilot read and write data in real systems through a typed tool catalog. With Onplana's public MCP server connected, you ask Claude "what's overdue in my Onplana?" and it returns the list, not a hallucinated guess. Then you say "mark Task A as done" and it does, against your real org, audited end to end.

Which AI clients work with Onplana?

Eight clients confirmed against the live MCP server at mcp.onplana.com/mcp: Claude Desktop (and Claude.ai web via Connectors Directory), ChatGPT custom connectors (Plus / Pro / Team / Enterprise / Edu), ChatGPT Codex CLI, Cursor, Gemini CLI, Gemini Code Assist (VS Code + JetBrains), GitHub Copilot in VS Code, and Grok. Any other client that speaks the MCP Streamable HTTP transport works identically — there is no per-client integration on our side.

How fast can I set this up?

Two minutes per client. Gemini CLI is a one-line install (`gemini extensions install https://github.com/Onplana/onplana-mcp-server`). Claude Desktop is a JSON snippet pasted into Settings → Developer → Edit Config and a restart. ChatGPT custom connectors take an OAuth-style account link plus selecting Onplana from the connector picker. The slowest path (which is still under five minutes) is signing up for an Onplana account and minting a personal access token.

Do I need a paid plan to use the MCP server?

No. The MCP server is available on every plan including Free. Tool availability is plan-gated — some tools (workflow automation, governance reviews) only appear when the calling org is on a plan that includes those features — but the core read-and-write surface (projects, tasks, sprints, milestones, comments, risks) is open to every account. The exact gate mapping mirrors the in-app feature gating, see the Pricing page for the full matrix.

Is my org data safe?

Every MCP call is scoped to a single org by the bearer token, audited in the AiOperation ledger (the same ledger the in-app AI assistant uses), and rate-limited per token. Tokens auth via either OAuth 2.1 (the recommended path for cloud agents like claude.ai's Connectors Directory flow) or a personal access token with the MCP_AGENT scope (for desktop clients). Tokens are revocable from Settings → Integrations → Connected Apps. Onplana never sees the AI client's training data, conversation history, or any context outside the typed tool calls.

What kind of prompts work well?

Direct, specific, action-oriented prompts. "What's overdue in Onplana?" works. "Show me sprints ending this week" works. "Create a task in the Onboarding project: draft welcome email sequence, due Friday, assign to me" works. Where MCP shines over generic AI is that the answer reflects your actual data, not a paraphrase or guess. Where it's weaker is open-ended planning ("design our Q3 roadmap") — for that, the in-app Onplana AI Assistant (which has memory + RAG over your full project history) is the better surface.

How does this compare to the in-app Onplana AI Assistant?

Different surfaces, same backend. The MCP server lets you do PM work without leaving your AI client of choice — handy when you're already in Claude reading a customer email or in Cursor writing code and a related project task lookup belongs in that flow. The in-app Onplana AI Assistant has memory mode, RAG over your wiki / comments / history, and the on-demand help catalog — better for sustained PM work and structured planning. Most teams end up using both: MCP for "quick lookups + drive-by mutations," in-app assistant for deeper work.

What if a prompt does the wrong thing?

Every mutating tool call writes a compensating action to the AiOperation ledger. One click in Settings → Activity reverses it, with the audit trail preserved. Preview-then-confirm is on by default for destructive operations (delete project, mass-update tasks). For read operations there's nothing to reverse — they just return data.

Can I use this with my own AI agent (not Claude / ChatGPT)?

Yes. The MCP server speaks the open Streamable HTTP transport from the MCP standard — any client that implements that protocol works against onplana with no per-client work on our side. If you're building a custom agent, see the engineering deep-dive at /mcp/how-we-built-it for the architecture notes and the public docs at /mcp for the tool catalog.

How is the search across my org indexed?

Hybrid BM25 + vector search. BM25 handles exact-term matches (a project named "Q3 Launch" surfaces for the query "Q3 Launch"); the vector layer handles semantic matches (a query about "blockers" finds risks and overdue dependencies even when the word "blocker" doesn't appear verbatim). The two scores are combined via Reciprocal Rank Fusion before returning the top results to the AI client. Indexes refresh continuously as you create and update content.

It combines enterprise-grade project management features like Gantt charts, dependencies, and portfolio management with a modern AI-powered and collaborative experience. It feels like a strong modern alternative for organizations moving away from Microsoft Project.

Waheed H., Manager · G2

Two minutes from here to "what's overdue?" answered by Claude.

Free plan covers the MCP server. PAT scope MCP_AGENT is the entry point. Hybrid search, audit ledger, undo, plan-aware tool gating — all included.

We use strictly-necessary cookies to operate this site (sign-in, anti-spam). With your consent, we also use Google Analytics 4 (anonymized IP) to understand which pages are useful. No ad tracking. See our Cookie Policy and Privacy Policy.