Connect Claude to Your Project Management Tool

Connecting Claude to a project management tool is a settings-panel task now, not an engineering project: generate a token or run one OAuth flow, and Claude can read your actual tasks, sprints, and risks, then draft updates from inside the conversation instead of you copying data back and forth by hand.

The direct answer: connect Claude to your project management tool over the Model Context Protocol (MCP), the open standard Anthropic released for exactly this. For Claude Desktop or Claude.ai, paste a personal access token or run through an OAuth connector; for Claude Code, one CLI command does it and opens a browser to authorize. What Claude can do after that is bounded by two things that matter more than the setup steps: the connecting user's own permissions, so it never sees more than that person could see in the app, and the vendor's tool catalog, since a delete action Claude can't call is a mistake it can't make.

How to Connect Claude to Your Project Management Tool

The setup differs slightly by which Claude client you're using, but the shape is the same everywhere: authorize once, then Claude has a standing connection until you revoke it.

1. Generate a credential. In Onplana, go to Integrations → AI agents and generate a connection token (Bearer personal access token, MCP_AGENT scope) or use the OAuth 2.1 flow if your client supports Custom Connectors.
2. Add the server to your Claude client. In Claude Desktop, that's a mcpServers entry in claude_desktop_config.json pointing at https://mcp.onplana.com/mcp with the token as a Bearer header. In Claude Code, it's one line: claude mcp add onplana --transport http https://mcp.onplana.com/mcp, which triggers OAuth in a browser on the first tool call instead of asking you to paste anything.
3. Restart and verify. Restart the client, then ask it something concrete: "list my projects" or "what's overdue on the Q3 rollout." If the answer matches what you'd see in the app, the connection is live.

The full walkthrough with exact file paths for macOS and Windows is at how to connect Onplana to Claude Desktop; the MCP server overview covers the same setup for other clients (Cursor, ChatGPT, Gemini) if Claude isn't the only assistant your team runs.

What Claude Can Actually Do Once It's Connected

The setup steps are the easy part. What determines whether the connection is safe to leave running is the scope of what's on the other end of it.

	What it means
Reads	List and search projects, tasks, sprints, risks, comments, and wiki content, the same records the authorizing user can already open in the app
Writes	Create and update tasks, comments, and status fields; the specific write tools available depend on the connecting user's role and plan
Never exposed	Delete actions. There are no delete_* tools in Onplana's MCP catalog; the furthest a write goes is a reversible status change, logged the same as a human edit
Bounded by	The authorizing user's own permissions, not a separate, broader "agent" identity; Claude cannot see a project that user couldn't open

That last row is the one worth checking with any vendor, not just Onplana. The security questions worth asking an AI agent vendor covers the same scoping question from the buyer's side: an integration that grants a connected assistant org-wide access by default, rather than the connecting user's own scope, turns one leaked token into an incident that touches everything instead of one project.

The diagram below shows the three properties that should hold true of any Claude connection you approve, regardless of vendor.

Why the Connection Limit Matters More Than It Looks

A plan-based cap on concurrent connections reads like a minor commercial detail until a token leaks or a trial account gets shared more widely than intended. A hard ceiling, set by an admin and enforced server-side, is what keeps that scenario contained to a small, known number of live connections instead of an unbounded one nobody's tracking. Onplana's caps run from 2 concurrent connections on Free and Starter up to 10 on Enterprise and unlimited on Enterprise Plus, and mutating actions carry a separate monthly cap on Free and Starter specifically, unlimited from Pro up, so a runaway script or an over-eager automation hits a wall before it becomes an incident rather than after.

That's a different constraint from what a lot of teams expect to be the limiting factor, which is usage cost. The cap here is about blast radius, not billing: how many live connections can exist and how many writes one can make in a month, independent of how much any individual conversation with Claude costs to run.

Claude Isn't the Only Client, But It's Usually the First

Most teams that connect an AI assistant to their project data start with whichever one they already use daily, and for a large share of PM buyers that's Claude. The setup pattern here, a scoped credential, a standard protocol, a capped connection, is the same one that should apply to ChatGPT, Cursor, and Gemini if your team runs more than one assistant, and the underlying protocol is the same across all of them: what MCP actually changes for a PM tool covers the mechanics in more depth if you're evaluating the standard itself rather than just wiring up Claude.

Connect Claude to Onplana in about 5 minutes Generate a scoped token, paste it into Claude Desktop's config (or run one command in Claude Code), and Claude can read and update your real projects from the conversation. → Read the full connect guide

More on how Onplana handles agent access, scope, and audit logging lives on the Onplana blog alongside this one.