Conclave

Conclave MCP is an MCP server that enables you to query multiple frontier LLM models, obtain diverse opinions, have peer-ranked evaluations, and receive a synthesized, consensus-based answer. It helps you surface alternatives, compare approaches, and make well-informed decisions in technical, business, creative, and research contexts.

How to use

You interact with Conclave MCP through an MCP client to run multi-model queries. Choose a usage pattern based on how deep you want the analysis.

• Quick brainstorming with conclave_quick to receive fast, diverse opinions from all configured models. - Ranked analysis with conclave_ranked to see which model performed best on your prompt. - Full synthesis with conclave_full to generate a final answer that blends multiple perspectives, including a chairman’s synthesis and an overall consensus level.

You can configure which models participate in the conclave, switch between tiered model sets, and adjust the chairman behavior. Use custom selections or default tier-based pools to steer cost and quality trade-offs. You will see options for tier, chairman presets, and consensus thresholds to tune how the final answer is produced.

How to install

# Prerequisites are assumed to be installed (Git, Python, and the runtime tooling shown below).

# Clone the repository
git clone https://github.com/stephenpeters/conclave-mcp.git
cd conclave-mcp

# Create .env file with your API key
echo "OPENROUTER_API_KEY=sk-or-v1-your-key-here" > .env

# Install dependencies or initialize runtime tooling
uv sync

Configuration and usage notes

The server supports both local (stdio) and remote (http) configurations. The standard local setup runs via a command such as uv run --directory /path/to/conclave-mcp python server.py, with the environment variable OPENROUTER_API_KEY provided to access OpenRouter models.

To incorporate Conclave MCP into Claude Desktop or Claude Code, use explicit stdio configurations that point to your local server. For example, you can run the server from a local directory and pass the API key via environment variables in the client configuration.

Security and maintenance

Keep your OpenRouter API key secure and do not commit it to version control. Rotate keys periodically and use separate keys for development and production where possible.

Monitor costs by using the available tools to estimate spend before running queries. The conclave supports tiered model sets to balance cost and performance.

Troubleshooting

If the server fails to start, verify that the environment variable OPENROUTER_API_KEY is set and that the path to the Conclave MCP directory is correct in your client configuration. Check that the required runtime (uv) is installed and available in your system PATH.

Example workflows

Custom conclave selection lets you pick specific models. Start with premium, standard, or budget pools, then fine-tune by designating a chairman or using a chairman preset. Run quick checks with conclave_quick, then escalate to conclave_ranked or conclave_full for deeper analysis and synthesis.

Available tools

conclave_quick

Fast parallel opinions (Stage 1 only). Queries all conclave models and returns individual responses.

conclave_ranked

Opinions with peer rankings (Stage 1 + 2). Shows which model performed best on this specific question.

conclave_full

Complete conclave with synthesis (all 3 stages). Includes consensus detection and chairman tiebreaker.

conclave_config

View current configuration: conclave members, chairman rotation status, consensus thresholds.

conclave_estimate

Estimate costs before running a query.

conclave_models

List all available models with selection numbers and tier groupings.

conclave_select

Create a custom conclave from model numbers; the first model becomes the chairman.

conclave_reset

Clear custom conclave selection and return to tier-based configuration.