Open Claude, go to Settings → Connectors, and you'll see a directory: Asana, Linear, Notion, a handful of others, plus an "Add custom connector" button for anything not listed. Click one, authorize it, and Claude starts answering questions about your actual tasks. It looks like Claude just became your project manager.
It didn't, quite. Claude's custom connectors link it to a project management tool's remote MCP server, so it can read — and for several tools, create or update — real tasks from inside a chat, no plugin store required. That's genuinely useful. But a connector gives Claude API-level access to your data, not a seat on your board with its own identity. Here's what a Claude custom connector for project management actually does, what it costs, and where the line to something more sits.
What is a custom connector, exactly?
A connector is a remote MCP (Model Context Protocol) server that Claude can call from inside a conversation. Anthropic and vendors publish pre-built ones you pick from a directory — Asana, Linear, and a growing list of others show up there today. A custom connector is the same mechanism pointed at a server that isn't in that directory: your own tool, an internal system, or a third-party server not yet listed.
Setting one up is the same either way: Settings → Connectors → Add custom connector, paste the server's URL, add OAuth credentials if it needs them, and authorize. No API keys to copy around, no separate plugin marketplace. Once it's connected, Claude can call whatever operations that server exposes, the same way it calls a pre-built one.
One thing worth correcting up front: several guides ranking for this topic say custom connectors require a paid Claude plan. That's not accurate as of this writing — free-tier accounts get one custom connector; Pro, Max, Team, and Enterprise plans remove that one-connector cap and add org-level controls. If you're on the free plan wondering whether you can even try this, the answer is yes, for one tool.
Which project management tools have a Claude connector today?
The pattern across the PM connectors that exist — Asana, Linear, monday.com-style tools, and similar — is the same shape: read tasks, projects, and status, and for most of them, create or update records too. What a specific connector can do depends on what that vendor built and how much of their own API they exposed through it; Claude doesn't add capabilities a tool's server doesn't have.
Write access is where it gets more deliberate. On Team and Enterprise plans, an admin can scope each connector's permissions into three buckets — Always allow, Needs approval, Blocked — separately for read and write actions. That's a real governance layer, not an afterthought: someone decided that letting an AI create tasks in your PM tool without a human glancing at it first is a choice, not a default.
What's the difference between a connector and a member of your team?
This is the part most explainers skip, and it's the one that matters most if you're actually trying to decide how AI should touch your project work.
A connector gives Claude a way to reach into your PM tool's data from a chat window. The result of that call lives in that conversation — whoever asked the question sees the answer. It doesn't automatically become something the rest of the team sees on the board itself: no avatar next to the task, no line in a shared activity feed saying an AI touched it, unless the underlying tool's own audit log happens to record API writes distinctly (most don't, by default — they attribute the change to whichever account authorized the connector).
That's a meaningfully different model from an AI that's a member of the board. Comuna's approach is the second one: you connect Claude the same way — Settings → Connectors, OAuth, no API keys — but then you add it to a specific board as a coworker with its own badge. Every card it touches, every comment it leaves, shows up in the shared activity feed with its name on it, visible to every human on that board, not just to the person who happened to be chatting with Claude when it ran. A connector answers one person's question. A board membership leaves a trail the whole team can see.
We wrote more about that distinction — three tiers, from assistant to automation to real membership — in what MCP means for project management; this is the same idea applied specifically to how Claude's connector directory works today.
Where connectors fall short right now
Being honest about the limits is what makes the rest of this worth trusting.
It's still pull, not push. A connector doesn't sit there watching your board. Claude reaches for it when a chat message — yours, or a scheduled prompt you configured — asks it to. Nothing runs continuously in the background; if you want a daily check, you have to set that up yourself as a scheduled prompt.
Visibility stops at the chat, by default. Unless the PM tool's own connector is built to log AI writes distinctly in its activity history, a teammate looking at the board later has no way to know a task was created or edited via a Claude connector rather than typed in directly. For a solo user that's a non-issue. For a team, it's the gap that turns "Claude helped" into "who changed this card and why."
Write access needs someone to actually configure it. The Always allow / Needs approval / Blocked controls are good design, but they're opt-in — an admin has to go set them per connector, per organization. Skip that step and you're relying on the vendor's defaults, which vary.
A connector is bounded by what the other tool's server exposes. If the PM tool only publishes five read-only operations and calls that "AI support," a Claude connector to it will only ever do five things. The protocol doesn't add capabilities the server doesn't have.
What this looks like for Comuna today
Comuna publishes its own MCP server — over 80 operations across cards, columns, comments, and notes — and Claude connects to it exactly the way this whole article describes: Settings → Connectors, OAuth, under two minutes, no API keys, and it works on the free Claude plan's one-connector slot. The difference isn't the connection mechanism; it's what happens after. You add the AI to a board as a coworker, it gets a name and a badge, and its actions land in the same activity stream your human teammates see — not a private answer in someone's chat. If it hits a decision it isn't sure about, it opens a small approval request instead of guessing. That's the board-membership model we think a connector is a good first step toward, not the finish line.
If you haven't set this up yet, the walkthrough is short: connect Claude to a board in 60 seconds.
FAQ
Do Claude's custom connectors require a paid plan?
No — free-tier Claude accounts can add one custom connector. Pro, Max, Team, and Enterprise plans remove that cap and add per-organization permission controls, but a paid plan isn't a hard requirement to try a single connector.
Can a Claude connector for project management create and update tasks, or only read them?
It depends on the tool's own MCP server, but most PM connectors in the directory today support both — reading tasks and status, and creating or updating records. On Team and Enterprise plans, an admin can restrict write access per connector (Always allow, Needs approval, or Blocked).
Is a Claude connector the same thing as an MCP server?
Not exactly. MCP is the open protocol; a connector is Claude's name for a specific remote MCP server you've authorized, whether it's one from Anthropic's directory or a custom one you point Claude at yourself. Every connector is backed by an MCP server, but "MCP server" is the broader, protocol-level term.
Does Comuna use a custom connector too?
Yes — Comuna's MCP server connects to Claude through the same Settings → Connectors flow described here. The difference is what happens after connecting: Claude joins a specific board as a named coworker, and its actions are visible to the whole team in the activity feed, not only inside the chat that triggered them.
Comuna is free forever — no credit card, bring your own AI. Spin up a workspace and try it.