MCP

Connect Claidex to Claude, OpenAI, VS Code, Cursor, and other Model Context Protocol clients.

Endpointhttps://app.claidex.com/api/mcp

npm package@claidex-mcp/claidex-mcp

Registry namecom.claidex/failure-intelligence

Quickstart

Use the remote HTTP endpoint when your client supports headers. Use the npm bridge when your client expects a local stdio server.

initialize

curl -sS https://app.claidex.com/api/mcp \
  -H "content-type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "initialize",
    "params": {
      "protocolVersion": "2025-11-25",
      "capabilities": {},
      "clientInfo": { "name": "precheck", "version": "1" }
    }
  }'

Claude

Claude Desktop works best through the npm stdio bridge. Claude Code can connect directly to the remote HTTP endpoint.

claude_desktop_config.json

{
  "mcpServers": {
    "claidex": {
      "command": "npx",
      "args": ["-y", "@claidex-mcp/claidex-mcp"],
      "env": {
        "CLAIDEX_MCP_URL": "https://app.claidex.com/api/mcp",
        "CLAIDEX_MCP_API_KEY": "claidex_sk_live_..."
      }
    }
  }
}

claude code

claude mcp add-json claidex '{
  "type": "http",
  "url": "https://app.claidex.com/api/mcp",
  "headers": {
    "Authorization": "Bearer ${CLAIDEX_MCP_API_KEY}"
  }
}'

OpenAI

Use the Responses API MCP tool with the Claidex remote endpoint. Limit allowed tools when a workflow needs a smaller tool surface.

responses api

import OpenAI from "openai";

const client = new OpenAI();

const response = await client.responses.create({
  model: "gpt-5",
  input: "Find terminated PIK3CA programs and summarize the translational risk.",
  tools: [
    {
      type: "mcp",
      server_label: "claidex",
      server_description: "Biomedical failure intelligence for claims, trials, targets, and preprints.",
      server_url: "https://app.claidex.com/api/mcp",
      authorization: process.env.CLAIDEX_MCP_API_KEY,
      allowed_tools: ["search", "fetch", "search_claims", "get_claim_content"],
      require_approval: "never"
    }
  ]
});

console.log(response.output_text);

Editors

Workspace-level configuration keeps the endpoint close to the project and stores secrets outside committed source.

.vscode/mcp.json

{
  "servers": {
    "claidex": {
      "type": "http",
      "url": "https://app.claidex.com/api/mcp",
      "headers": {
        "Authorization": "Bearer ${input:claidex-api-key}"
      }
    }
  },
  "inputs": [
    {
      "id": "claidex-api-key",
      "type": "promptString",
      "description": "Claidex API key",
      "password": true
    }
  ]
}

cursor mcp.json

{
  "mcpServers": {
    "claidex": {
      "url": "https://app.claidex.com/api/mcp",
      "headers": {
        "Authorization": "Bearer ${CLAIDEX_MCP_API_KEY}"
      }
    }
  }
}

Capabilities

Claidex exposes tools for action, resources for context, and prompts for reusable biomedical analysis workflows.

Tools	24 model-callable tools for claims, trials, target risk, literature, statistics, computation, and retrieval.
Resources	Platform status, tool inventory, prompt catalog, metrics, and security model references.
Prompts	Reusable workflows for failure post-mortems, target dossiers, preprint audits, and phase transition analysis.
Compatibility	Native `search` and `fetch` tools for OpenAI and ChatGPT-style retrieval flows.

Registry

Use these exact values when publishing or validating the Claidex MCP registry listing.

server.json

{
  "name": "com.claidex/failure-intelligence",
  "title": "Claidex MCP",
  "version": "1.0.0",
  "remotes": [
    {
      "transportType": "streamable-http",
      "url": "https://app.claidex.com/api/mcp"
    }
  ],
  "packages": [
    {
      "registryType": "npm",
      "identifier": "@claidex-mcp/claidex-mcp",
      "version": "1.0.0"
    }
  ]
}