Sensor Tower MCP Server

An MCP server that lets agents use Sensor Tower APIs for ads, market, and utility data—no custom HTTP clients needed.

Features

• FastMCP-based MCP server with structured tool registration.
• Coverage for App Analysis, Market Analysis, Store Marketing, Usage Intelligence, and utility endpoints.
• Built-in input normalization for common parameter issues (e.g., ad network aliases).
• Bundled OpenAPI specifications (swaggerdocs/) for reference and validation.

Requirements

• Python 3.10+
• Sensor Tower API token (SENSOR_TOWER_API_TOKEN environment variable)
• (Optional) Docker for container-based deployment

Quick Start

1. Install dependencies and the package:

   uv sync  # or: pip install -e .[test]
   

2. Export your API token:

   export SENSOR_TOWER_API_TOKEN="st_xxxxxxxxx"
   

3. Launch the MCP server:

   python -m sensortower_mcp.server
   

   The FastMCP CLI will expose the registered tools to your orchestrator.

Client setup (Cursor, Claude, Docker)

Cursor

• Settings → MCP → Add Server → Command:
  • Command: python
  • Arguments: -m sensortower_mcp.server
  • Env: set SENSOR_TOWER_API_TOKEN

Example JSON (if Cursor asks for a block):

{
  "name": "sensortower",
  "command": "python",
  "args": ["-m", "sensortower_mcp.server"],
  "env": { "SENSOR_TOWER_API_TOKEN": "st_xxxxxxxxx" }
}

Claude (Desktop/Web with MCP)

• Add a custom MCP server in Settings → Integrations → Model Context Protocol.
• Use the same command/args/env as above.

Minimal config snippet some clients accept:

{
  "mcpServers": {
    "sensortower": {
      "command": "python",
      "args": ["-m", "sensortower_mcp.server"],
      "env": { "SENSOR_TOWER_API_TOKEN": "st_xxxxxxxxx" }
    }
  }
}

Docker

Build and run the server container locally:

docker build -t sensortower-mcp .
docker run --rm \
  -e SENSOR_TOWER_API_TOKEN=st_xxxxxxxxx \
  -p 8666:8666 \
  sensortower-mcp

Or with Compose (uses docker-compose.yml):

docker compose up -d

Then point your MCP client at the local command as usual, or at http://localhost:8666 if your client supports HTTP MCP transport.

HTTP Invocation Shortcut

When exercising the server over HTTP without a JSON-RPC session, target the legacy gateway exposed at /legacy/tools/invoke:

curl \
  -H 'Content-Type: application/json' \
  -H 'Accept: application/json, text/event-stream' \
  -X POST "http://localhost:8666/legacy/tools/invoke" \
  -d '{
        "tool": "get_creatives",
        "arguments": {
          "os": "ios",
          "app_ids": "835599320",
          "start_date": "2023-01-01",
          "end_date": "2023-01-31",
          "countries": "US",
          "networks": "Instagram,Unity",
          "ad_types": "video"
        }
      }'

The endpoint always responds with { "tool": "<name>", "result": <structured payload> }. The combined Accept header is mandatory because FastMCP's HTTP transport performs content negotiation that requires both application/json and text/event-stream to be present; omitting it can yield 406 errors in some clients. The server automatically widens overly strict Accept values from known MCP clients, but external tooling should set the header explicitly to avoid regression issues.

Usage Example

Calling the creatives tool from FastMCP requires the ad_types parameter, which Sensor Tower marks as mandatory:

from fastmcp import FastMCP
from sensortower_mcp.server import build_server

mcp = FastMCP("sensortower")
server = build_server(api_token="st_xxxxxxxxx")
server.register(mcp)

result = mcp.invoke_tool(
    "get_creatives",
    {
        "os": "ios",
        "app_ids": "284882215",
        "start_date": "2024-01-01",
        "countries": "US",
        "networks": "Instagram",
        "ad_types": "video"
    }
)

Testing

Install the testing extras and exercise the suites using uv (preferred) or the active virtualenv:

uv sync --extra test
uv run pytest tests/test_result_normalization.py  # fast structural sanity check
uv run pytest                                     # full offline suite

Live API exercises are marked with @pytest.mark.live_api and require a valid SENSOR_TOWER_API_TOKEN. Execute only that suite with:

uv run pytest -m live_api

The MCP server must be running locally (see Quick Start), and HTTP callers should send Accept: application/json, text/event-stream—or use the legacy gateway, which automatically normalizes the header when known MCP clients connect.

Manual smoke scripts that hit the deployed package or raw MCP transports have been moved to manual/. They are not part of automated CI but remain available for ad-hoc verification:

• manual/search_entities_fix.py – verifies the published package’s search_entities helper.
• manual/search_entities_mcp.py – end-to-end JSON-RPC smoke test for search_entities.

Run them with python manual/<script>.py; each script prints a summary and exits 0/1 accordingly.

See docs/testing.md for the full test matrix, CI recommendations, and troubleshooting tips.

Release Highlights

• 1.2.10: Expands tool docstrings with metadata-driven usage hints so MCP clients surface examples and notes without extra manual edits.
• 1.2.1: Ensures get_creatives forwards the required ad_types query parameter, preventing 422 responses from Sensor Tower. Adds regression coverage so missing parameters are detected before release.

Contributing

Issues and merge requests are welcome. Please include tests for bug fixes and keep API examples in the README and README-pypi.md synchronized.