NUSK MCP
Your project plan, inside your AI agent.
Ask Claude Code what’s blocked and update the sprint, without leaving your terminal. NUSK MCP is built into every workspace: connect Claude Code, Codex, Cursor or any other MCP client, and it works with your permissions, nothing more.
What your agent can do
Nothing to install. Your agent connects to your workspace over HTTPS and acts as you, with your project permissions.
Find work
List your projects, see your open work across all of them, search and filter work items, and read an item in full with its comments.
Plan
Create work items, change status, owner, priority, dates and estimates, and move items between sprints and the backlog.
Collaborate
Comment on work items and turn meeting action items into tasks.
Track risk
Read the RAID log and sprint progress, and add new risks.
Set it up
- Copy your server URL. It is your workspace address followed by
/api/mcp, for examplehttps://nusk.dev/your-workspace/api/mcp. The API tokens page in your profile shows it ready to copy. - Add it to your agent using one of the snippets below. Replace
your-workspacewith your own workspace name. - Sign in. Your agent opens a NUSK page in your browser. Check the name, choose Read only if the agent should only look or Read and change to let it make changes, then click Allow. No token to copy.
Tools that cannot open a browser sign-in (scripts, CI, some older clients) can use a personal access token instead: create one in Profile → API tokens and send it as Authorization: Bearer nusk_.... The token snippets are further down.
Claude Code
Run this once in a terminal, then type /mcp in Claude Code, pick nuskpm and choose Authenticate to sign in.
claude mcp add --transport http nuskpm https://nusk.dev/your-workspace/api/mcp
With a personal access token instead (no browser sign-in):
claude mcp add --transport http nuskpm https://nusk.dev/your-workspace/api/mcp --header "Authorization: Bearer nusk_..."
Codex
Codex uses a personal access token. Add this to ~/.codex/config.toml, then set NUSKPM_TOKEN to your token in your shell.
[mcp_servers.nuskpm] url = "https://nusk.dev/your-workspace/api/mcp" bearer_token_env_var = "NUSKPM_TOKEN"
Cursor, Windsurf and other MCP clients
Add this to the client's MCP configuration file, for example .cursor/mcp.json. The client offers to sign you in the first time it connects.
{
"mcpServers": {
"nuskpm": {
"url": "https://nusk.dev/your-workspace/api/mcp"
}
}
}If a client cannot sign in through the browser, add "headers": { "Authorization": "Bearer nusk_..." } with a personal access token.
Then ask your agent something like “What's blocking release 2 in NUSK?”, “What's on my plate in NUSK this week?” or “Create a task in project AB to add a logout button and assign it to me.”
Safe by design
The agent can never do more than you can.
- Your permissions — every request is checked against your role in each project.
- Read-only access — if you choose Read only when you sign in (or use a read-only token), the agent only gets the read tools.
- Short-lived sign-in — the agent’s access renews itself every hour and stops working if you end it.
- You approve changes — MCP clients ask you before each tool call that changes data.
- Full history — changes made by an agent are marked “via AI agent (MCP)” in revision history and can be reverted.
- Revoke any time — delete the agent’s entry (named “OAuth: …”) or a token in Profile → API tokens and it loses access at once.
Questions
Do I need to download anything?
No. The MCP server is part of every NUSK workspace. You only add its address to your agent and sign in.
Does it work with NUSK On-Premise?
Yes. Use your own server's address followed by /api/mcp, for example https://pm.example.com/api/mcp.
Can I add it to claude.ai or ChatGPT on the web?
NUSK now supports the standard OAuth sign-in that web connectors use, so you can add the server address as a custom connector and sign in. It also works with Claude Code, Cursor and VS Code through the same sign-in, and with Codex using a token. Connector support varies by app and plan.
Do I still need a token?
Not for agents that can open a browser sign-in. Tokens remain for scripts, CI and clients that only accept a header.
Is my data sent to the AI provider?
Only what your agent reads through the tools you allow, and only to the AI provider your agent already uses.