Connect SciRouter to Claude via MCP: Complete Setup Guide

What Is MCP and Why It Matters for Science

The Model Context Protocol (MCP) is an open standard that lets AI assistants like Claude discover and call external tools automatically. Instead of writing custom integration code, you configure an MCP server and the AI agent can see what tools are available, understand their parameters, and invoke them as needed during a conversation.

For scientific computing, this is transformative. A researcher can ask Claude to fold a protein, calculate molecular properties, or dock a compound — and Claude will automatically call the right SciRouter API endpoints, interpret the results, and present them in context. No SDK, no scripts, no boilerplate.

How SciRouter's MCP Server Works

SciRouter runs an MCP server that exposes its full catalog of scientific computing tools over the SSE (Server-Sent Events) transport protocol. When Claude Desktop connects to the server, it receives a list of available tools with their names, descriptions, and parameter schemas. Claude can then call any tool during a conversation by sending a function call request back through the same connection.

• Discovery: Claude sees all available science tools and their descriptions automatically
• Type safety: Parameter schemas ensure Claude sends valid inputs (sequences, SMILES strings, etc.)
• Async handling: The MCP server manages job polling for GPU-intensive operations like protein folding
• Authentication: Your API key is configured once and used for all tool calls

Step 1: Get Your SciRouter API Key

If you do not already have a SciRouter account, sign up for free. Navigate to the API Keys section of your dashboard and create a new key. Copy the key — it will start with sk-sci- and you will only see it once.

Step 2: Configure MCP in Claude Desktop

Open your Claude Desktop configuration file. On macOS this is located at ~/Library/Application Support/Claude/claude_desktop_config.json. On Windows it is at %APPDATA%\Claude\claude_desktop_config.json. Add the SciRouter MCP server entry:

{
  "mcpServers": {
    "scirouter": {
      "url": "https://mcp.scirouter.ai/sse",
      "headers": {
        "Authorization": "Bearer sk-sci-your-api-key-here"
      }
    }
  }
}

Save the file and restart Claude Desktop. The SciRouter tools should now appear in Claude's tool list — you can verify by clicking the hammer icon in the chat input area.

Step 3: Verify Claude Can See the Tools

Open a new conversation in Claude Desktop and ask it what science tools are available. Claude should list the SciRouter tools it discovered via MCP. Here is an example of what a conversation looks like:

You: What scientific computing tools do you have access to?

Claude: I have access to several scientific computing tools through SciRouter:

- **ESMFold**: Predict protein 3D structure from amino acid sequence
- **DiffDock**: AI-powered molecular docking
- **Molecular Properties**: Calculate drug-likeness metrics from SMILES
- **Molecule Similarity**: Tanimoto similarity between compounds
- **Format Conversion**: Convert between SMILES, InChI, and MOL formats

Would you like me to use any of these tools?

You: Fold this protein sequence: MVLSPADKTNVKAAWGKVGAHAGEYGAEALERMFLSFPTTKTYFPHFDLSH

Claude: I'll fold that sequence using ESMFold.

[Calls SciRouter ESMFold tool]

The structure prediction is complete. Here are the results:
- Mean pLDDT score: 87.3 (high confidence)
- The predicted structure shows a well-folded alpha-helical domain
- This appears to be a hemoglobin subunit based on the fold topology

Available Tools via MCP

The following tools are currently available through SciRouter's MCP server. Each tool includes full parameter documentation that Claude uses to construct valid requests.

• ESMFold — Predict protein 3D structure from amino acid sequence
• Molecular Properties — Calculate molecular weight, logP, TPSA, H-bond donors/acceptors from SMILES
• DiffDock — Dock small molecules against protein structures using diffusion models

The full and growing list of tools is available on the SciRouter tools catalog.

MCP vs Direct API vs SDK: When to Use Which

SciRouter offers three ways to access its tools. Each serves a different use case:

MCP (This Guide)

Best for interactive exploration with AI assistants. Zero code required. Claude discovers tools automatically and handles all API mechanics. Ideal for researchers who want to query science tools conversationally.

Direct REST API

Best for production integrations and custom pipelines. Full control over request parameters, error handling, and retry logic. Use when you need deterministic behavior or are building a web application.

Python SDK

Best for scripting and notebook workflows. Provides typed Python objects and handles authentication, polling, and serialization. Use when writing Python scripts or Jupyter notebooks for reproducible analysis.

• Choose MCP when you want Claude or another AI to orchestrate tool calls interactively
• Choose the REST API when building applications or services that call SciRouter programmatically
• Choose the SDK when writing Python scripts, notebooks, or LangChain agents

Next Steps

With MCP configured, Claude can now act as your scientific computing assistant. Ask it to fold proteins, screen compounds, or compare molecular properties — all through natural conversation.

To build more complex automated workflows, see our guide on using SciRouter with LangChain for AI drug discovery. Or explore the full tools catalog to see everything available through the API.