Legion MCP Server

MCP-native LLM councils for debates, juries, blind panels, refinement, and custom deliberation.

Local serverstdio

What is the Legion MCP Server?

Legion is an MCP integration. MCP-native LLM councils for debates, juries, blind panels, refinement, and custom deliberation.

Packages and local setup

The published npm package is legion-mcp, version 0.8.0. Its declared MCP transport is stdio.

The package declares MCP_TRANSPORT (required): Run the installed package over MCP stdio (the binary defaults to HTTP).; DEFAULT_BASE_URL: Default OpenAI Responses-compatible API root for model files that omit baseUrl.; DEFAULT_API_KEY: Default provider or gateway key for model files that omit apiKey.; ALLOW_NO_MODELS (required): Boot without any config/models/.json. The package ships only key-free examples, which the scanner ignores, so this must stay true until a real model file exists beside the working directory.; OPENAI_API_KEY: Optional. A model file's apiKey may be "env:VAR" to read the key from an environment variable named VAR at startup; name those variables here as needed. The bundled gpt example reads this one.; MAX_ROUNDS: Maximum discussion rounds accepted by a council run.; TOKEN_BUDGET: Optional soft cumulative token budget for a council run.; DYNAMIC_ROLES: Allow callers to define ad-hoc council roles inline.; LOG_LEVEL: Logging verbosity. Logs are written to stderr for stdio safety.; ASYNC_TOOLS: Run councils in the background: council tools return a job handle and poll/cancel tools are exposed. Jobs live in process memory only.; TRUST_PROXY_AUTH: Bind async jobs to the caller named by X-MS-CLIENT-PRINCIPAL- headers. Only safe behind an ingress that strips client-supplied copies..

Before connecting

Check the maintainer’s instructions for account access, authentication and the supported client configuration. Once connected, inspect the tools your client receives and try a small task that matches the server’s stated purpose. The registry description does not include a verified, complete tool list.

Project and source

View the publisher’s registry entry · Project repository. This profile uses version 0.8.0 from the registry. Registration identifies the publisher namespace; it does not establish that this is a first-party integration for every service it connects.

Browse related developer tools MCP servers for tools that cover the same kind of work.

Configuration

VariableDescriptionRequired
MCP_TRANSPORTRun the installed package over MCP stdio (the binary defaults to HTTP).Yes
DEFAULT_BASE_URLDefault OpenAI Responses-compatible API root for model files that omit baseUrl.Optional
DEFAULT_API_KEYDefault provider or gateway key for model files that omit apiKey.Optional
ALLOW_NO_MODELSBoot without any config/models/.json. The package ships only key-free examples, which the scanner ignores, so this must stay true until a real model file exists beside the working directory.Yes
OPENAI_API_KEYOptional. A model file's apiKey may be "env:VAR" to read the key from an environment variable named VAR at startup; name those variables here as needed. The bundled gpt example reads this one.Optional
MAX_ROUNDSMaximum discussion rounds accepted by a council run.Optional
TOKEN_BUDGETOptional soft cumulative token budget for a council run.Optional
DYNAMIC_ROLESAllow callers to define ad-hoc council roles inline.Optional
LOG_LEVELLogging verbosity. Logs are written to stderr for stdio safety.Optional
ASYNC_TOOLSRun councils in the background: council tools return a job handle and poll/cancel tools are exposed. Jobs live in process memory only.Optional
TRUST_PROXY_AUTHBind async jobs to the caller named by X-MS-CLIENT-PRINCIPAL- headers. Only safe behind an ingress that strips client-supplied copies.Optional