Plurality
Uses SQLite checkpoints to retain private run history, including complete responses, partial failures, usage, and finish reasons, with JSON and Markdown export.
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
Plurality AI
Think through a decision with several AI models. Plurality gathers independent perspectives, lets them respond to one another, and returns a decision memo with the disagreements and complete transcript preserved.
Use it as a personal MCP server in Codex, Claude Desktop or another MCP host, with your own OpenRouter key. The repository also includes the Nuxt website, shared deliberation engine, tests and reusable agent skills.
Personal MCP quick start
Install Node.js 22 or later and get an OpenRouter API key. From a copy of this repository:
npm ci
node scripts/mcp-direct.mjs --helpSave your key outside the repository in a private file, for example ~/.config/plurality/openrouter-api-key, readable only by you (chmod 600). Configure your MCP host with absolute paths:
{
"mcpServers": {
"plurality": {
"command": "/absolute/path/to/node",
"args": ["/absolute/path/to/plurality-ai-mcp/scripts/mcp-direct.mjs"],
"env": {
"OPENROUTER_API_KEY_FILE": "/absolute/path/to/private/openrouter-api-key"
}
}
}
}Reconnect the host, then ask: “Use Plurality to review this plan against my constraints.”
The MCP needs no hosted app, subscription, Stripe account or Postgres database. OpenRouter charges for model usage. The launcher reads its process environment; it does not load .env automatically.
Full setup and Codex configuration
Related MCP server: tokonomix-council-mcp
What it does
Eight methods: Quick Take, Consensus, Debate, Red Team, Peer Review, Forecast, Brainstorm and Oracle voting.
Named persona teams: choose complementary roles and models, including multiple roles using the same model.
Preview and background work: inspect the roster and call bounds, start a saved run, and retrieve it by ID.
Private history: SQLite checkpoints retain complete responses, partial failures, usage and finish reasons. Export JSON or Markdown and record usefulness feedback.
Bounded recovery: token-limit continuations are capped; interrupted and failed work stays inspectable. Network failures do not automatically replay paid calls.
Results lead with a memo; transcripts remain available through plurality_result and plurality_export. Runs live in ~/.local/share/plurality by default, configurable with PLURALITY_DATA_DIR. This archive is private by file permissions, but is not encrypted at rest. Prompts and supplied evidence are sent to OpenRouter and the selected model providers.
Model agreement is not factual verification. Same-model personas are correlated, Oracle percentages are vote shares or ratings, and forecasts are not calibrated probabilities. A panel can cost more without producing a better answer; the included comparison harness helps evaluate that tradeoff.
Development and checks
npm test
npm run build
npm run mcp:eval -- --fixture # offline synthetic comparison
npm run mcp:eval # preview the paid experiment; no generationLive experiments require an explicit --execute and incur provider charges. Keep generated responses and personal data outside source control.
Directory | Contents |
| Shared models, prompts, parsers and deliberation utilities |
| Personal MCP planning, execution and SQLite persistence |
| Personal stdio MCP entry point |
| Review, persona-team and Oracle agent skills |
| Reproducible comparison code and synthetic cases |
| Nuxt website and authenticated hosted API |
| Offline unit, workflow and recovery checks |
Self-hosting the website
The website is a separate setup with Supabase authentication, Postgres persistence and optional commercial integrations. It retains the original branding, pricing screens and deployment assumptions. The personal MCP is the simplest way to start.
cp .env.example .env
# Configure your own services and operator details.
npm run devWebsite setup and deployment checklist · Billing runbook
Contributing and security
See CONTRIBUTING.md for local checks and contribution scope, and SECURITY.md for secret handling and private vulnerability reporting. Both package manifests remain private: true to prevent accidental npm publication; this does not restrict using or modifying the source.
License
Apache License 2.0. Copyright © 2026 Thor Matthiasson. See NOTICE. Dependencies retain their own licenses.
This server cannot be deployed
Maintenance
Related MCP Connectors
Put a decision to a council of top AI models and get one brief: verdict, splits, risks, next steps.
Convene a panel of expert AI personas to debate any decision from every side.
mumo is a remote MCP server for multi-model deliberation. Your agent sends a question to models from different labs — Claude, GPT, Gemini, Grok, DeepSeek, Kimi, and more — and gets back their full responses plus typed cross-model reactions. The participating models react to each other directly and explain, in their own words, what they agree with, challenge, or want to explore further. Agents can run a deliberation with `create_deliberation`, wait for results with `wait_for_round`, and steer follow-up rounds with `append_round` using structured signals like KEEP, EXPLORE, CHALLENGE, CORE, and SHIFT. Built for architecture, plan/spec review, strategy, and pre-launch pressure tests. Free tier available. API key required.
Five AI models debate your decision blind, then give one recommendation and the strongest objection.
Related MCP Servers
- AlicenseAqualityDmaintenanceProvides access to multiple frontier LLM models (GPT, Claude, Gemini, Grok, DeepSeek) for consulting a "conclave" of AI perspectives, enabling peer-ranked evaluations and synthesized consensus answers for important decisions.81MIT

tokonomix-council-mcpofficial
AlicenseAqualityBmaintenanceEnables multi-model consensus decision-making for high-stakes AI decisions, using independent expert models and a judge to surface disagreements and ground decisions.11164 npm1MIT- AlicenseNot gradedqualityAmaintenanceEnables orchestrating a panel of frontier AI models that independently propose, critique, revise, and converge on a single agreed answer to a problem, using existing vendor subscriptions or API keys.13 npm1MIT
- AlicenseNot gradedqualityCmaintenanceEnables multi-model deliberation by sending questions to various AI models and aggregating their responses and typed cross-model reactions, facilitating agreement, challenge, and exploration for tasks like architecture review and strategy.MIT