PrimeOutpost MCP server

Ask your AI assistant about your security program and let it read the answer straight from PrimeOutpost: the roadmap, the risk register, the decision log, playbooks, framework coverage, maturity scores and what is in flight this week.

The server runs locally, is read-only, and can never change anything in PrimeOutpost. A personal access token belongs to exactly one organization, so the assistant sees that organization and nothing else.

Requirements

• uv installed, which provides the uvx command.
• A personal access token from Settings -> Security in PrimeOutpost.

Usage

### Claude Code

claude mcp add primeoutpost -e PRIMEOUTPOST_TOKEN=po_pat_... -- uvx primeoutpost-mcp

Claude Desktop and Cursor

Add this to your MCP configuration file:

{
  "mcpServers": {
    "primeoutpost": {
      "command": "uvx",
      "args": ["primeoutpost-mcp"],
      "env": {
        "PRIMEOUTPOST_TOKEN": "po_pat_..."
      }
    }
  }
}

Configuration

Tools

References

Entities are named by short references: RM-012 is a roadmap item, RISK-004 a risk, DEC-003 a decision and PB-002 a playbook. Numbers are per organization. Use them in your prompt exactly as they appear in the app. The assistant passes one to get_item to read it in full, or a batch to resolve_refs to see what a list of references points at. Case and padding do not matter, so rm-12 finds RM-012.

Security

Your personal access token is scoped to a single organization and carries your own role, so the assistant can read exactly what you can read and no more. Tokens expire, and you can revoke one at any time from Settings -> Security. The server runs on your machine, talks only to your PrimeOutpost API, and sends nothing anywhere else. Every tool is a read: there is no tool that creates, edits or deletes. What your assistant does with the data once it has read it is between you and whoever provides that assistant.