Model Context Protocol

Standpulse for AI coding agents

Standpulse runs an MCP server, so an agent working in your codebase can keep the board honest while it works: pick up the next task, claim it, comment progress, raise a blocker when it is stuck, and close the task with a summary when it is done.

• Tasks and blockers. List, create, start, complete, assign and comment; raise and resolve blockers.
• Docs. Read project docs for context, and write new ones.
• Planner and standup. Plan work into a day, and summarise what moved.
• Ask the project. Grounded answers from docs, tasks and blockers, with sources.

Get a key

1. A team lead opens Settings → Agent access in Standpulse.
2. Choose a name, the member the key acts as, its abilities and, optionally, specific projects and an expiry date.
3. Copy the key. It looks like sp_mcp_…​.… and is shown once; only a hash is stored. If you lose it, create a new one.

Revoke a key from the same screen at any time; it stops working immediately.

Connect a client

Claude Code

claude mcp add --transport http standpulse https://www.standpulse.com/mcp/pm \
  --header "Authorization: Bearer sp_mcp_xxxx.yyyy"

Check the connection with /mcp, then ask the agent to call whoami.

Other MCP clients

Any client that supports remote HTTP servers with custom headers. Most accept a JSON block like this:

{
    "mcpServers": {
        "standpulse": {
            "type": "http",
            "url": "https://www.standpulse.com/mcp/pm",
            "headers": {
                "Authorization": "Bearer sp_mcp_xxxx.yyyy"
            }
        }
    }
}

Keep the key out of version control. Prefer your client's environment-variable substitution where it has one.

Permissions

Two layers decide what an agent can do, and the key can only ever narrow the member:

• The member the key acts as. Their role decides what is visible, exactly as in the app. A key acting as an intern sees only what that intern sees.
• The key carries abilities and an optional project allow-list.

Everything the agent writes is attributed to that member and marked as coming from MCP in the activity log. For an agent that should see a whole project, act as a team lead or project manager; for a clean audit trail, create a dedicated member for the agent.

Tools

Every tool returns structured content plus a one-line text summary. Ids are integers.

Orientation

Tasks

Plans

Blockers

Docs

Planner

Team and reporting

Prompts

Prompts are workflows the server ships: each returns the live board context it needs plus a step-by-step procedure, so every connected agent runs the same playbook. In Claude Code they appear as slash commands — /mcp__standpulse__plan — and other clients list them under prompts/list. They run in sequence: brainstorm writes a doc, plan turns it into tasks, parallel_work works them.

A prompt needs the same abilities as the tools it leans on: brainstorm needs docs:read, plan tasks:read, parallel_work tasks:write. Its context never includes a project the key is not scoped to.

Limits

• Paging. List tools return at most 100 records per call and a next_cursor; pass it back as cursor for the next page.
• Rate limit. 120 requests per minute per key.
• ask_project needs the knowledge add-on on your plan. whoami reports which projects have it.

Troubleshooting