For AI agents: a documentation index is available at /llms.txt. A markdown version of this page is available at /guide/mcp/quickstart.md.
Quickstart
Connect a client over OAuth, approve the scopes, and make your first assistant-driven Orangescrum call.
About five minutes. There is no API key to fetch first. You sign in through the browser.
One prerequisite
MCP has to be enabled for your workspace before any sign-in can succeed. If you have not had that confirmed, start at Authentication.
Check the server is reachable
This call has no token, so a healthy server answers
401, that is the result you want. It proves you can reach the endpoint and that it is speaking OAuth:curl -i -sS -X POST https://v4-api.orangescrum.com/mcp/partner \ -H 'Content-Type: application/json' \ -d '{"jsonrpc":"2.0","method":"ping","id":1}'Look for
WWW-Authenticate: Bearer realm="MCP"in the response headers. No response at all means a network or proxy problem, fix that here rather than in a client config.Add the server to your client
Every client needs the same one thing: the URL
https://v4-api.orangescrum.com/mcp/partner. Claude.ai and ChatGPT take it in their connector UI with no file to edit. For Claude Code:claude mcp add --transport http orangescrum https://v4-api.orangescrum.com/mcp/partnerSee Client setup for Cursor, VS Code, Codex CLI and Claude Desktop.
Sign in and pick a workspace
The client opens a browser window the first time it connects. Sign in with your Orangescrum account or Google, then choose the workspace this assistant should work in.
The grant is bound to that one workspace, connect a second time if you need a second one.
Approve the scopes
The consent screen lists what the client is asking for, grouped by area. Approve only what you want it to do.
Starting read-only is the sensible default:
mcp:useplus the.readscopes for the areas you care about. You can re-authorize with write access once you trust it.Confirm the tools appeared
Ask the assistant what Orangescrum tools it has. It should describe tools across projects, tasks, sprints, test management and more. If it says it has none, jump to Troubleshooting.
Try a real request
Start with a read so you can see the shape of the data before anything writes:
List my Orangescrum projects and tell me which ones have tasks due this week.
Then try a write once you are comfortable, and have granted a write scope:
Create a task in Website Redesign called "Audit meta descriptions", assign it to me, and set the due date to next Friday.
#Useful first prompts
"What's overdue across my projects, grouped by assignee?"
"Run the daily_standup prompt."
"What's in the active sprint, and what's still sitting in the backlog with no estimate?"
"List the test cases for the checkout scenario and flag any with no steps."
"Log 2 hours against OS-431 for yesterday with the note 'schema migration'."
"Search our uploaded documents for the signed SOW with Acme."
Check writes before you trust them
Assistants can misread which project or task you meant, especially when names are similar. For the first few sessions, confirm each write in Orangescrum before moving on, and keep the grant read-only until you have.