Client Setup

AI Workflow MCP uses Streamable HTTP MCP at:

https://mcp.aiworkflowmcp.com/mcp

Use the same endpoint in every client. Add an API key when the client supports headers.

Environment Variable

Store the key outside the config file when possible:

export AIWORKFLOWMCP_API_KEY="ak_..."

Never commit real keys into a repository.

Claude Code

claude mcp add --transport http aiworkflowmcp https://mcp.aiworkflowmcp.com/mcp

If your Claude Code version supports request headers for HTTP MCP servers, add:

Authorization: Bearer ${AIWORKFLOWMCP_API_KEY}

If headers are not supported in your current client build, public calls still work. Use REST with the bearer header for authenticated smoke tests until the client supports it.

Cursor

Add an HTTP MCP server entry:

{
  "mcpServers": {
    "aiworkflowmcp": {
      "url": "https://mcp.aiworkflowmcp.com/mcp",
      "headers": {
        "Authorization": "Bearer ${AIWORKFLOWMCP_API_KEY}"
      }
    }
  }
}

Restart the editor after changing MCP configuration.

Cline

Use the same HTTP server shape:

{
  "mcpServers": {
    "aiworkflowmcp": {
      "url": "https://mcp.aiworkflowmcp.com/mcp",
      "headers": {
        "Authorization": "Bearer ${AIWORKFLOWMCP_API_KEY}"
      }
    }
  }
}

Then ask the agent to list available tools. A healthy client should show get_ai_model, get_coding_tool, search_official_docs, search_skills, search_cli, search_mcp, search_frameworks, route_framework, and get_framework.

Windsurf, Codex, OpenCode, And Generic Clients

Use the generic MCP configuration if the client supports Streamable HTTP:

{
  "name": "aiworkflowmcp",
  "transport": {
    "type": "http",
    "url": "https://mcp.aiworkflowmcp.com/mcp",
    "headers": {
      "Authorization": "Bearer ${AIWORKFLOWMCP_API_KEY}"
    }
  }
}

If the client expects a simpler object, keep only:

{
  "url": "https://mcp.aiworkflowmcp.com/mcp",
  "headers": {
    "Authorization": "Bearer ${AIWORKFLOWMCP_API_KEY}"
  }
}

Verify The Connection

Use tools/list first because it does not require tool arguments:

curl -fsS https://mcp.aiworkflowmcp.com/mcp \
  -H "content-type: application/json" \
  -H "Authorization: Bearer ${AIWORKFLOWMCP_API_KEY}" \
  --data '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "tools/list"
  }'

Then call a deterministic read tool:

curl -fsS https://mcp.aiworkflowmcp.com/mcp \
  -H "content-type: application/json" \
  -H "Authorization: Bearer ${AIWORKFLOWMCP_API_KEY}" \
  --data '{
    "jsonrpc": "2.0",
    "id": 2,
    "method": "tools/call",
    "params": {
      "name": "search_official_docs",
      "arguments": {
        "query": "Claude Code MCP HTTP configuration",
        "fresh_only": true,
        "limit": 3
      }
    }
  }'

Agent Prompt Contract

Tell the agent:

Use AI Workflow MCP before answering current AI coding tool, model, MCP server, or agent framework questions. Query in concise English. Prefer fresh_only=true for source-sensitive answers. Cite evidence.source_url when making product or pricing claims.

Common Setup Mistakes

Using /v1/... as the MCP URL. MCP clients must use /mcp.
Sending a key without the Bearer prefix.
Passing an empty argument such as model_id: "".
Expecting localized tool names. Tool names and JSON fields stay English.
Treating public unauthenticated access as production quota. Use an API key for real workflows.

Client Setup

On this page