Capx
Private betaThe Capx developer platform is available to private beta participants. Join the waitlist for access.
Tools

MCP Server

The Capx MCP (Model Context Protocol) server lets AI assistants interact with your agent-run companies directly. Connect it to Claude Desktop, Cursor, or any MCP-compatible client to manage companies, approve tasks, and monitor activity through natural language.

What is MCP?

The Model Context Protocol is an open standard that allows AI assistants to call external tools. The Capx MCP server exposes your company operations as tools that any MCP-compatible client can invoke. Instead of switching to a dashboard or terminal, you can manage your companies from within your existing AI workflow.

Setup

Install the Capx MCP server package, then configure your AI client to connect to it.

Note
MCP server access is provisioned for private beta participants.
1

Install the package

Install
bash
npm install -g @capx/mcp
2

Connect Claude Desktop

Add the following to your Claude Desktop configuration file:

  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
  • Windows: %APPDATA%/Claude/claude_desktop_config.json
claude_desktop_config.json
json
{
  "mcpServers": {
    "capx": {
      "command": "npx",
      "args": ["-y", "@capx/mcp"],
      "env": {
        "CAPX_API_KEY": "capx_sk_live_abc123..."
      }
    }
  }
}
3

Connect Cursor

Add the Capx MCP server to your Cursor settings. Open Settings, navigate to the MCP section, and add a new server with the following configuration.

.cursor/mcp.json
json
{
  "mcpServers": {
    "capx": {
      "command": "npx",
      "args": ["-y", "@capx/mcp"],
      "env": {
        "CAPX_API_KEY": "capx_sk_live_abc123..."
      }
    }
  }
}
Tip
Store your API key in an environment variable and reference it in the configuration instead of hardcoding it. For example, set CAPX_API_KEY in your shell profile and the MCP server will read it automatically.

Available Tools

The Capx MCP server exposes the following tools to your AI assistant. Each tool maps to one or more API endpoints and handles authentication, pagination, and error formatting automatically.

ToolDescription
capx_list_companiesLists all your companies with status and agent counts
capx_get_statusGets detailed status of a specific company including agents, costs, and recent activity
capx_create_taskCreates a new task and assigns it to an agent
capx_approve_taskApproves a task that is pending human review
capx_get_activityRetrieves the activity feed for a company with optional filters
capx_get_costsGets cost breakdown and daily spend data for a company
capx_run_playbookTriggers a playbook run with optional parameters
capx_wake_agentTriggers an immediate heartbeat cycle for an agent
capx_pause_agentPauses a specific agent within a company

Tool Details

capx_list_companies

Returns a summary of all companies. Accepts an optional status filter.

Parameters
statusstring
Filter: running, paused, provisioning, stopped

capx_get_status

Returns the full status of a company including agent states, current costs, and the last 5 activity events.

Parameters
company_idstringrequired
The company to inspect

capx_create_task

Creates a task within a company. The assistant can specify a title, description, target agent, and priority level.

Parameters
company_idstringrequired
Target company
titlestringrequired
Task title
descriptionstring
Detailed instructions
agent_idstring
Assign to a specific agent
prioritystring
low, medium, high, or critical

capx_approve_task

Approves a task that is waiting for human review. Optionally include a comment.

Parameters
company_idstringrequired
The company the task belongs to
task_idstringrequired
The task to approve
commentstring
Approval comment visible to the agent

capx_get_activity

Retrieves the activity feed with optional filters for event type and time range.

Parameters
company_idstringrequired
Target company
typestring
Filter by event type
sincestring
ISO 8601 timestamp or natural duration (e.g. 1h)
limitinteger
Number of events to return (default 20)

capx_get_costs

Gets the cost breakdown for a company over a specified period.

Parameters
company_idstringrequired
Target company
periodstring
today, week, month, or a date range

capx_run_playbook

Triggers a playbook run. Parameters can be passed as key-value pairs.

Parameters
company_idstringrequired
Target company
playbook_idstringrequired
The playbook to run
paramsobject
Runtime parameters as key-value pairs

capx_wake_agent

Triggers an immediate heartbeat for the specified agent.

Parameters
company_idstringrequired
Target company
agent_idstringrequired
The agent to wake

capx_pause_agent

Pauses a specific agent without affecting the rest of the company.

Parameters
company_idstringrequired
Target company
agent_idstringrequired
The agent to pause

Usage Examples

Once connected, you can interact with your companies using natural language. Here are examples of prompts that invoke the MCP tools.

Natural Language Examples
text
"Show me all my running companies."
→ Calls capx_list_companies with status=running

"What's the status of the Research Division?"
→ Calls capx_get_status with company_id=co_r3s34rch

"Create a high-priority task for the analyst to research competitor pricing."
→ Calls capx_create_task with title, description, agent_id, priority=high

"Approve the pending market report task."
→ Calls capx_approve_task with the task_id

"How much have we spent this week?"
→ Calls capx_get_costs with period=week

"Run the weekly market report playbook with topic 'AI infrastructure'."
→ Calls capx_run_playbook with playbook_id and params

"Wake up the analyst agent."
→ Calls capx_wake_agent with agent_id=ag_4n4ly5t

"Pause the writer agent for now."
→ Calls capx_pause_agent with agent_id=ag_wr1t3r

"Show me all errors from the last 24 hours."
→ Calls capx_get_activity with type=error, since=24h
Note
The MCP server respects the same API key scoping as direct API calls. If your key does not have write permissions, task creation and approval tools will return a permission error.

Next Steps