MCP Tools

POST JSON-RPC 2.0 requests to:

https://mcp.aiworkflowmcp.com/mcp

Initialize

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "initialize",
  "params": {
    "protocolVersion": "2025-06-18",
    "capabilities": {},
    "clientInfo": {
      "name": "example-client",
      "version": "1.0.0"
    }
  }
}

The server returns serverInfo.name: "aiworkflowmcp-cloudflare-v2" and a tools capability object.

List Tools

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

Call A Tool

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "get_ai_model",
    "arguments": {
      "model_id": "gpt-5"
    }
  }
}

Tools

• get_ai_model(model_id): model pricing, context, modalities, limits, provenance, and evidence.
• get_coding_tool(tool_id): install, command surface, supported models, environment variables, and source evidence.
• search_official_docs(query, tool_id?, include_markdown?, fresh_only?): official docs with optional R2 Markdown.
• search_skills(task, harness?, fresh_only?): official and major-vendor agent Skills.
• search_cli(task, tool_id?, include_docs?, fresh_only?): CLI command-surface facts.
• search_mcp(task, official_only?, include_docs?, fresh_only?): official and major-vendor MCP servers.
• search_frameworks(query, domain?, include_candidates?, limit?): thinking framework search.
• route_framework(task, domain?, include_candidates?): route a user task to an operating framework.
• get_framework(framework_id): retrieve a published framework card.

Tool Selection

Use get_ai_model when the user names a specific model id such as gpt-5 or claude-opus-4-7.

Use get_coding_tool when the user names a tool id such as claude-code, openai-codex-cli, cursor, or windsurf.

Use search tools when the user describes a task and the agent needs discovery. Keep the query concise and English-first.

Use framework tools when the user asks for a decision path, operating checklist, risk review, product strategy, or agent workflow routing.

Response Contract

Successful tools/call responses use this shape:

{
  "jsonrpc": "2.0",
  "id": 1,
  "result": {
    "content": [
      {
        "type": "text",
        "text": "{ ... }"
      }
    ],
    "structuredContent": {
      "id": "gpt-5",
      "evidence": {
        "source_url": "https://example.com",
        "key": "evidence/model/gpt-5/hash.json",
        "content_hash": "hash",
        "stale": false
      }
    }
  }
}

Agents should read structuredContent for deterministic behavior and reserve content for user-facing summaries.

Freshness

Use fresh_only=true for commercial paths that must reject stale facts. Health degrades when stale public facts, source errors, or quality gate failures are visible.

Limits

tools/call is rate-limited. Public calls use IP-minute scope. Authenticated calls use API-key-minute scope.

Current production limits are exposed by /health:

public: 120 requests/minute
authenticated: 600 requests/minute

Unsupported JSON-RPC Behavior

Batch requests are not supported. Notifications and accepted MCP signal payloads return 202 with an empty body. Unknown methods return JSON-RPC -32601.