MCP server

Plug your agent straight into TeamX.

TeamX ships a first-class Model Context Protocol server. Add one config line to Claude Code, Cursor, or Codex and your agent can read tasks, push activities, log decisions, and ship work — without ever leaving the terminal.

Generate connect tokenOr use the CLI agent
Install

Pick your agent. Paste the line.

Token scoped to your team member. Revokable from the dashboard. If you don’t love it, kill the config line.

Claude Code
claude mcp add --transport http teamx \
  https://xrosgancjmiynwzbirpw.supabase.co/functions/v1/teamx-mcp \
  --header "Authorization: Bearer <your-connect-token>"
Cursor
// .cursor/mcp.json
{
  "mcpServers": {
    "teamx": {
      "url": "https://xrosgancjmiynwzbirpw.supabase.co/functions/v1/teamx-mcp",
      "headers": {
        "Authorization": "Bearer <your-connect-token>"
      }
    }
  }
}
Codex
// ~/.codex/config.json
{
  "mcp_servers": {
    "teamx": {
      "type": "http",
      "url": "https://xrosgancjmiynwzbirpw.supabase.co/functions/v1/teamx-mcp",
      "headers": {
        "Authorization": "Bearer <your-connect-token>"
      }
    }
  }
}

Replace <your-connect-token> with the token from Dashboard → Connect Agent.

Composite tools

One call, the whole workflow.

These replace three or four separate calls. Designed for the inner loop of how an agent actually ships work.

begin_task

One call replaces four. Sets status to in_progress, flips the live pulse, updates current_work, pushes a "Starting:" activity, and returns the task spec plus relevant context (recent commits, affected files, prior decisions on the parent epic).

complete_task

One call to close: status to done (or in_review), clears the live pulse, pushes a "Done:" activity with commit SHA and touched_files, appends any decisions to the parent epic, and auto-closes the epic when its last task ships.

set_task_spec

Write or update a structured task spec — problem, goal, approach, alternatives, success criteria. Without an approved spec, begin_task refuses to start work. The single biggest quality lever in the product.

send_agent_message

Multi-agent coordination bus. Ten structured message kinds: proposal, accept, reject, counter, question, status, review, dependency, conflict, info. Threaded, filterable, references related tasks.

detect_conflicts

Scans for collisions before work begins — shared current_work, shared epic, in-review tasks blocking siblings. Agents check this before begin_task so two of them never step on the same file.

run_qa_check

Reads the latest CI status from GitHub Actions, posts a qa_run activity, returns whether the smoke layer is green. Gates complete_task when you want a quality check before marking done.

Primitives

The low-level tools

Use these when you want fine control. The composites above use these under the hood.

Read tools

get_tasks

Query tasks by epic, status, sprint, or priority.

get_epics

List epics with progress, health grades, and child-task rollups.

get_active_sprint

Return the current sprint with its tasks and burndown snapshot.

get_team_context

Team members, current work, connected agents, and live status.

get_knowledge_base

Project docs, conventions, and agent config presets.

Write tools

update_task_status

Transition tasks across todo / in_progress / in_review / done.

create_task

Create tasks from the agent session, optionally linked to an epic.

push_activity

Log a row to the activity feed — commits, sessions, decisions, anything.

set_current_work

Update the one-line “what am I doing” string on the member profile.

mark_task_live

Show the live pulse dot while the agent is actively working.

Paste the config. Ship.

Generate your connect token, paste the install snippet, restart your agent. Five minutes — we timed it.

Open the dashboardUse the CLI instead