multipov-mcp-server

MCP server for multipov.ai — a multi-perspective AI review platform that runs any document, plan, or code diff past a panel of LLM-powered personas and synthesizes their findings. Exposes 14 tools for reviews, rewrites, and persona browsing from any MCP-aware client (Claude Code, Claude Desktop, Cursor, Codex, Zed, Raycast).

This package is a thin stdio-to-HTTP proxy. It speaks MCP over stdio to your local client and forwards every request to the hosted endpoint at https://multipov.ai/mcp. No business logic runs locally — the persona catalog, the review engine, and the rewrite engine all live server-side.

Install

npx multipov-mcp-server

The server runs over stdio. Your MCP client spawns it as a child process and talks to it through pipes.

Authenticate

1. Sign in at multipov.ai.
2. Open /settings/api-keys → Generate.
3. Copy the mpov_live_... token. It is shown once — store it in your password manager.

Register with Claude Code

claude mcp add multipov npx -y multipov-mcp-server --env MULTIPOV_API_KEY=mpov_live_YOUR_TOKEN_HERE

Or, by hand, in ~/.claude.json (or your project-local .mcp.json):

{
  "mcpServers": {
    "multipov": {
      "command": "npx",
      "args": ["-y", "multipov-mcp-server"],
      "env": {
        "MULTIPOV_API_KEY": "mpov_live_YOUR_TOKEN_HERE"
      }
    }
  }
}

Restart the client and run /mcp to verify the multipov server is connected.

Register with Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or the equivalent on your OS and add the same snippet as above under mcpServers.

Register with Cursor

Paste the same snippet into ~/.cursor/mcp.json under mcpServers.

What's inside

14 tools across four groups:

• Personas — list_personas, get_persona, recommend_personas
• Reviews — submit_review, get_review_status, get_review_report, cancel_review, list_my_reviews
• Rewrites — submit_rewrite, get_rewrite_status, get_rewrite_result
• Specialized reviews — submit_plan_review, submit_pipeline_review, submit_codebase_review

All tools share the same daily quota as the multipov.ai web UI. See the full reference at capitalthought/multipov-docs.

Why a proxy instead of connecting to the HTTP endpoint directly?

You can connect directly to https://multipov.ai/mcp if your client supports streamable-HTTP transport:

claude mcp add --transport http multipov https://multipov.ai/mcp \
  --header "Authorization: Bearer mpov_live_YOUR_TOKEN_HERE"

This package exists because not every MCP client supports HTTP transport yet (many still default to stdio), and because the claude mcp add npx -y ... UX is consistent across most of the MCP ecosystem. Under the hood this package is a ~150-line proxy that uses the official @modelcontextprotocol/sdk streamable-HTTP client to forward tool calls to the hosted endpoint.

Configuration

| Env var | Required | Default | Notes | | --- | --- | --- | --- | | MULTIPOV_API_KEY | yes | — | Get one at /settings/api-keys. | | MULTIPOV_BASE_URL | no | https://multipov.ai | Override for local development. |

The API key is never logged, never written to disk, never bundled in the package. It is only ever read from the environment at startup and sent in the Authorization header to the hosted endpoint.

Privacy

Submitted content is processed server-side by Anthropic, OpenAI, Google, and xAI per multipov.ai/privacy. This proxy does not add any telemetry, does not phone home, and does not cache your content locally.

License

MIT © Capital Thought