Build Your Agent on Claude Managed Agents
TL;DR
In about 15 minutes, create an agent in the Claude console, give it a unique, job-based name, choose a model, and define its responsibilities and team conventions. Add Wrike, then set its URL to https://mcp.wrike.com/v2 in Raw view - the directory’s default URL is outdated. Keep tool permissions on auto, securely copy your API key, and confirm the agent is Active. Next, connect it to Wrike as a Workmate; nothing is connected or visible to colleagues yet.
| Availability: Business, Pinnacle, Apex. ; Unavailability: Free, Team; |
Note
This is an experimental feature released in Wrike Labs. It is subject to change and may not be available to all users.
Overview
Everything in this article happens in the Claude console, on your own Anthropic account. Nothing here touches Wrike, and nothing you do here is visible to your colleagues. It covers only the choices that decide how your Workmate behaves once it reaches Wrike - Anthropic's quickstart is the full account of building an agent.
Creating a Workmate will take about fifteen minutes.
How to Build Your Agent
1. In the Claude console, create a new agent.
2. Name it after the job it does. This name is what your team sees in Wrike when they mention it, so "Project Manager" tells a colleague what to expect where "test-agent-3" does not.
3. Choose a model. This is the biggest single influence on the quality of the Workmate's judgment, and on what it costs you. A smaller model costs less per job and is noticeably weaker at the calls that matter - deciding which risk is the serious one, or noticing that two dates in a brief contradict each other.
4. Write its instructions. This is what makes it a project manager rather than a general assistant. Cover:
- The kinds of job it takes on, and how it decides which one it has been asked for.
- How it should read your account before forming a view.
- What it should treat as evidence, and what it should admit it cannot know.
- Your team's own conventions - how you name projects, what your statuses mean, who decides what.
5. Add Wrike as a connection, then switch the agent's configuration to the Raw view and set the Wrike address to https://mcp.wrike.com/v2. That address is how your agent reaches your Wrike work - Wrike's MCP server documentation describes what it can read and change.
6. Leave the tool permission setting on auto. This lets the agent decide for itself which actions it can take directly and which ones need your approval.
7. Create an API key and copy it. You will paste it into Wrike in the next article.
Important
The Claude console's Wrike directory entry points are at an older address, and an agent configured from that entry cannot reach your work. Set the address in the Raw view, as described in step 5. After you save, the ordinary view still displays the directory's address - the value in the Raw view is the one that is saved and the one your agent uses.
Note
Wrike takes the agent's name exactly as you wrote it, and the name cannot be changed in Wrike afterwards. Give each agent a name no other agent on your account shares, or your colleagues will see two Workmates they cannot tell apart in the mention menu.
Note
You do not sign in to Wrike here yet. Your Workmate gets its own Wrike identity when you connect it, and that identity is what decides what it can see.
Tip
Treat the API key the way you would treat a password. It gives Wrike access to your Anthropic account, and anyone holding it has the same access.
You are ready to continue when the console shows your agent as Active, with the model you chose, your instructions in its configuration, and url: https://mcp.wrike.com/v2 under mcp_servers in the Raw view.
What's Next?
Was this article helpful?
Thank you for your feedback.
Thank you for your feedback.
Please sign in to leave a comment.