Cohere MCP Server

A Type 3 DAuth MCP server for Cohere AI API. Provides chat, embeddings, reranking, classification, and summarization using Cohere's AI models.

Features

• Chat — Conversational AI similar to chat completions
• Embed — Text embeddings with multiple input types
• Rerank — Rerank documents by relevance
• Classify — Classify text into categories
• Summarize — Text summarization
• List Models — View available Cohere models

Authentication

This server uses Type 3 DAuth (Dedalus Auth) — your API key is encrypted client-side and decrypted in a secure Dedalus enclave.

Get Your Cohere API Key

1. Go to https://cohere.ai/AI-API
2. Sign up and get an API key
3. Copy the key

Installation

git clone https://github.com/dedalus-labs/cohere-mcp.git
cd cohere-mcp
pip install -e .
cp .env.example .env
# Edit .env and add COHERE_API_KEY

Available Tools

cohere_chat

Chat with Cohere AI.

cohere_chat(
    message="Hello!",
    model="command",
    temperature=0.7,
    max_tokens=500,
)

cohere_embed

Generate text embeddings.

cohere_embed(
    texts=["Hello world", "This is a test"],
    model="embed-english-v3.0",
    input_type="search_document",
)

cohere_rerank

Rerank documents by relevance.

cohere_rerank(
    query="What is AI?",
    documents=["AI is great", "Python is a language"],
    model="rerank-english-v3.0",
    top_n=10,
)

cohere_classify

Classify texts.

cohere_classify(
    texts=["This is great", "This is terrible"],
    model="embed-english-v3.0",
)

cohere_summarize

Summarize text.

cohere_summarize(
    text="Long text to summarize...",
    model="command",
    length="medium",
    format="paragraph",
)

cohere_list_models

List available Cohere models.

cohere_list_models()

Pricing

Cohere offers pay-per-token pricing. Check https://cohere.ai/pricing for details.

Deploy to Dedalus

1. Push to GitHub (public repo)
2. Go to https://www.dedaluslabs.ai/dashboard
3. Add Server → Connect GitHub repo
4. Set COHERE_API_KEY as Required Credential
5. Deploy

License

MIT