Vol. III · Multi-Model Reasoning · ed. 2026
Plurality
pp. xlv — l
Article the Ninth · The MCP server

Plurality, in your editor.

A multi-model deliberation server accessible directly from Claude Code, Claude Desktop, Cursor, Windsurf, and any MCP-compatible client. Five tools, thirty frontier models, one credit balance shared with the web app.

§ I — Connect, in three steps
  1. 01
    Mint an API key.

    Visit your billing page and create a key. The full secret is shown only once — copy it now. Open billing →

  2. 02
    Add the server to your client.

    Paste this into the MCP config file your client reads — for Claude Desktop it is ~/Library/Application Support/Claude/claude_desktop_config.json on macOS. Restart the client.

    MCP config
    {
      "mcpServers": {
        "plurality": {
          "type": "http",
          "url": "https://pluralityai.com/api/mcp",
          "headers": {
            "Authorization": "Bearer plk_…your-key-here…"
          }
        }
      }
    }
  3. 03
    Use it.

    Ask the assistant a hard question and explicitly invoke a tool — e.g. "Run plurality_consensus on whether to deprecate our REST API for GraphQL". Every call deducts from the same credit balance as the web app and saves to your archive.

§ II — The five tools

Each tool is the same deliberation pipeline the web app runs, exposed as an MCP tool. Pass a prompt (or target for Red Team) and an optional list of model IDs.

plurality_quick_takecf. Quick Take
Fast multi-model take. Each model answers independently, then a synthesizer weaves them into one essay.
plurality_consensuscf. Consensus
Multi-round deliberation that critiques and revises until the panel converges on the strongest synthesized answer.
plurality_debatecf. Debate
Adversarial debate that surfaces every angle of a contested question, with a judge weighing the strongest arguments.
plurality_forecastcf. Forecast
Calibrated probability estimate from a panel of forecasters, log-odds aggregated, with key assumptions surfaced.
plurality_red_teamcf. Red Team
Stress-test an idea, proposal, or pitch. Adversarial reviewers extract every flaw ranked by severity.
§ III — Of what is included
01
30 frontier models.
Claude, GPT, Gemini, Llama, Grok, DeepSeek, Qwen, Mistral, and more — orchestrated as a panel, not a single voice.
02
One credit balance.
MCP calls debit the same monthly credit balance as the web app. No separate billing.
03
Every call archived.
Each deliberation is saved to your archive with full transcript, model votes, and timing — replayable from the web app.
04
Works everywhere.
Connect from Claude Code, Claude Desktop, Cursor, Windsurf, or any MCP-compatible client that supports HTTP transport.
Ready to connect

Sign in to begin →