GitHub - agoda-com/api-agent: Universal MCP server for GraphQL/REST APIs

API Agent

Turn any API into an MCP server. Query in English. Get results—even when the API can't.

Point at any GraphQL or REST API. Ask questions in natural language. The agent fetches data, stores it in DuckDB, and runs SQL post-processing. Rankings, filters, JOINs work even if the API doesn't support them.

What Makes It Different

🎯 Zero config. No custom MCP code per API. Point at a GraphQL endpoint or OpenAPI spec — schema introspected automatically.

✨ SQL post-processing. API returns 10,000 unsorted rows? Agent ranks top 10. No GROUP BY? Agent aggregates. Need JOINs across endpoints? Agent combines.

🔒 Safe by default. Read-only. Mutations blocked unless explicitly allowed.

🧠 Recipe learning. Successful queries become cached pipelines. Reuse instantly without LLM reasoning.

Quick Start

1. Run (choose one):

# Direct run (no clone needed)
OPENAI_API_KEY=your_key uvx --from git+https://github.com/agoda-com/api-agent api-agent

# Or clone & run
git clone https://github.com/agoda-com/api-agent.git && cd api-agent
uv sync && OPENAI_API_KEY=your_key uv run api-agent

# Or Docker
git clone https://github.com/agoda-com/api-agent
docker build -t api-agent .
docker run -p 3000:3000 -e OPENAI_API_KEY=your_key api-agent

2. Add to any MCP client:

{
  "mcpServers": {
    "rickandmorty": {
      "url": "http://localhost:3000/mcp",
      "headers": {
        "X-Target-URL": "https://rickandmortyapi.com/graphql",
        "X-API-Type": "graphql"
      }
    }
  }
}

3. Ask questions:

• "Show characters from Earth, only alive ones, group by species"
• "Top 10 characters by episode count"
• "Compare alive vs dead by species, only species with 10+ characters"

That's it. Agent introspects schema, generates queries, runs SQL post-processing.

More Examples

REST API (Petstore):

{
  "mcpServers": {
    "petstore": {
      "url": "http://localhost:3000/mcp",
      "headers": {
        "X-Target-URL": "https://petstore3.swagger.io/api/v3/openapi.json",
        "X-API-Type": "rest"
      }
    }
  }
}

Your own API with auth:

{
  "mcpServers": {
    "myapi": {
      "url": "http://localhost:3000/mcp",
      "headers": {
        "X-Target-URL": "https://api.example.com/graphql",
        "X-API-Type": "graphql",
        "X-Target-Headers": "{\"Authorization\": \"Bearer YOUR_TOKEN\"}"
      }
    }
  }
}

Reference

Headers

MCP Tools

Core tools (2 per API):

Tool names auto-generated from URL (e.g., example_query). Override with X-API-Name.

Recipe tools (dynamic, added as recipes are learned):

Cached pipelines, no LLM reasoning. Appear after successful queries. Clients notified via tools/list_changed.

Configuration

How It Works

sequenceDiagram
    participant U as User
    participant M as MCP Server
    participant A as Agent
    participant G as Target API

    U->>M: Question + Headers
    M->>G: Schema introspection
    G-->>M: Schema
    M->>A: Schema + question
    A->>G: API call
    G-->>A: Data → stored in DuckDB
    A->>A: SQL post-processing
    A-->>M: Summary
    M-->>U: {ok, data, queries[]}

Architecture

flowchart TB
    subgraph Client["MCP Client"]
        H["Headers: X-Target-URL, X-API-Type"]
    end

    subgraph MCP["MCP Server (FastMCP)"]
        Q["{prefix}_query"]
        E["{prefix}_execute"]
        R["r_{recipe} (dynamic)"]
    end

    subgraph Agent["Agents (OpenAI Agents SDK)"]
        GA["GraphQL Agent"]
        RA["REST Agent"]
    end

    subgraph Exec["Executors"]
        HTTP["HTTP Client"]
        Duck["DuckDB"]
    end

    Client -->|NL + headers| MCP
    Q -->|graphql| GA
    Q -->|rest| RA
    E --> HTTP
    R -->|"no LLM"| HTTP
    R --> Duck
    GA --> HTTP
    RA --> HTTP
    GA --> Duck
    RA --> Duck
    HTTP --> API[Target API]

Stack: FastMCP • OpenAI Agents SDK • DuckDB

Recipe Learning

Agent learns reusable patterns from successful queries:

1. Executes — API calls + SQL via LLM reasoning
2. Extracts — LLM converts trace into parameterized template
3. Caches — Stores recipe keyed by (API, schema hash)
4. Exposes — Recipe becomes MCP tool (r_{name}) callable without LLM

flowchart LR
    subgraph First["First Query via {prefix}_query"]
        Q1["'Top 5 users by age'"]
        A1["Agent reasons"]
        E1["API + SQL"]
        R1["Recipe extracted"]
    end

    subgraph Tools["MCP Tools"]
        T["r_get_top_users<br/>params: {limit}"]
    end

    subgraph Reuse["Direct Call"]
        Q2["r_get_top_users({limit: 10})"]
        X["Execute directly"]
    end

    Q1 --> A1 --> E1 --> R1 --> T
    Q2 --> T --> X

Recipes auto-expire on schema changes. Disable with API_AGENT_ENABLE_RECIPES=false.

Development

git clone https://github.com/agoda-com/api-agent.git
cd api-agent
uv sync --group dev
uv run pytest tests/ -v      # Tests
uv run ruff check api_agent/  # Lint
uv run ty check               # Type check

Observability

Set OTEL_EXPORTER_OTLP_ENDPOINT to enable OpenTelemetry tracing. Works with Jaeger, Zipkin, Grafana Tempo, Arize Phoenix.