Chain Of Recursive Thoughts (cort) Mcp Server

@KunihiroS

About Chain Of Recursive Thoughts (cort) Mcp Server

This is a Chain-of-Recursive-Thoughts (CORT) MCP server.

Config

Add this server to your MCP-compatible client using the configuration below.

{
  "mcpServers": {
    "CoRT-chain-of-recursive-thinking": {
      "command": "pipx",
      "args": [
        "run",
        "cort-mcp",
        "--log=off"
      ],
      "env": {
        "OPENAI_API_KEY": "{apikey}",
        "OPENROUTER_API_KEY": "{apikey}"
      }
    }
  }
}

Tools

No tools detected

We auto-extract tools from the README. The maintainer can list them under a ## Tools heading to populate this section.

Overview

What is Chain Of Recursive Thoughts (cort) Mcp Server?

Chain Of Recursive Thoughts (cort) Mcp Server is a Model Context Protocol server that implements the Chain-of-Recursive-Thoughts method, making an AI think harder by making it argue with itself repeatedly. It integrates with OpenRouter and optionally OpenAI APIs, and is intended for use with MCP hosts such as Roo Code or Cline.

How to use Chain Of Recursive Thoughts (cort) Mcp Server?

Configure it in your MCP host using pipx run cort-mcp with environment variables OPENAI_API_KEY and OPENROUTER_API_KEY. Optional logging flags --log=on/off and --logfile can be set. A 300-second timeout is recommended because responses may take longer than expected.

Key features of Chain Of Recursive Thoughts (cort) Mcp Server

CoRT method: AI repeatedly argues with itself to deepen reasoning.
Multi LLM inference: randomly selects different model/provider per alternative.
Enhanced evaluation prompt that asks AI to explain its reasoning.
Tools: {toolname}.simple, .details, .mixed.llm, .neweval.
Fallback processing: auto-retries with OpenAI if OpenRouter call fails (conditions apply).

Use cases of Chain Of Recursive Thoughts (cort) Mcp Server

Enhancing AI reasoning for complex or ambiguous problem-solving tasks.
Exploring diverse perspectives by mixing multiple LLM providers.
Debugging reasoning flows with detailed response history logging.
Implementing recursive self-critique in MCP‑based AI agents.

FAQ from Chain Of Recursive Thoughts (cort) Mcp Server

How does the CoRT process work?

The server determines 1–5 thinking rounds, creates an initial response, then in each round generates three alternatives at temperatures 0.7, 0.8, 0.9 and evaluates them at temperature 0.2 to select the best response.

What API keys are required?

OPENROUTER_API_KEY is required to use OpenRouter. OPENAI_API_KEY is needed only when using OpenAI models or utilizing the fallback feature.

What logging options are available?

Set --log=off to disable all logging, or --log=on --logfile=/absolute/path to enable logging to a file. Both flags are required when logging is enabled.

What models are used in multi‑LLM mode?

A fixed list includes gpt-4.1-nano, meta-llama/llama-4-scout:free, google/gemini-2.0-flash-exp:free, mistralai/mistral-small-3.1-24b-instruct:free, meta-llama/llama-3.2-3b-instruct:free, and thudm/glm-4-9b:free.

What happens if an API call fails?

If the first API call to a non‑OpenAI provider fails and OPENAI_API_KEY is set, the server automatically retries with the default OpenAI model. Otherwise, the original error is returned.

Frequently asked questions

How does the CoRT process work?

What API keys are required?

`OPENROUTER_API_KEY` is required to use OpenRouter. `OPENAI_API_KEY` is needed only when using OpenAI models or utilizing the fallback feature.

What logging options are available?

Set `--log=off` to disable all logging, or `--log=on --logfile=/absolute/path` to enable logging to a file. Both flags are required when logging is enabled.

What models are used in multi‑LLM mode?

A fixed list includes `gpt-4.1-nano`, `meta-llama/llama-4-scout:free`, `google/gemini-2.0-flash-exp:free`, `mistralai/mistral-small-3.1-24b-instruct:free`, `meta-llama/llama-3.2-3b-instruct:free`, and `thudm/glm-4-9b:free`.

What happens if an API call fails?

If the first API call to a non‑OpenAI provider fails and `OPENAI_API_KEY` is set, the server automatically retries with the default OpenAI model. Otherwise, the original error is returned.