mcp-hevo-service

A read-only MCP server for the Hevo Data public API (v2.0). Self-hosted: your API credentials stay on your machine and never reach a third party. Exposes monitoring/inspection tools over pipelines, objects, destinations, models and workflows. Every tool maps to a GET endpoint, so the server cannot mutate Hevo state.

Quick start

git clone <this-repo>
cd mcp-hevo-service
pip install -r requirements.txt
cp .env.template .env   # then fill in your credentials

Configuration

Generate an API Key + Secret in Hevo (Account Settings > API) and set them in .env:

Variable	Description
HEVO_API_KEY	API key (used as the Basic-auth username).
HEVO_API_SECRET	API secret (used as the Basic-auth password).
HEVO_REGION	One of us, eu, au, in. Selects the base URL https://<region>.hevodata.com/api/public/v2.0.

Verify

python test_connection.py

Confirms the credentials authenticate against GET /pipelines. No MCP client needed.

Register in an MCP client

{
  "mcpServers": {
    "hevo": {
      "command": "/path/to/python3",
      "args": ["/path/to/mcp-hevo-service/server.py"]
    }
  }
}

Tools

All tools are read-only (GET).

Tool	Hevo endpoint
list_pipelines	GET /pipelines
get_pipeline	GET /pipelines/{id}
get_pipeline_schedule	GET /pipelines/{id}/schedule
list_pipeline_objects	GET /pipelines/{id}/objects
get_pipeline_object_stats	GET /pipelines/{id}/objects/{object_name}/stats
list_destinations	GET /destinations
get_destination	GET /destinations/{id}
get_destination_table_stats	GET /destinations/{id}/tables/{table_name}/stats
list_models	GET /models
get_model	GET /models/{id}
list_workflows	GET /workflows
get_workflow	GET /workflows/{id}

Security

• Read-only by construction: the client only issues GET requests and no write tool is registered. There is no way to start/pause/edit a pipeline through this server.
• Credentials are read from .env (gitignored) and sent as HTTP Basic auth directly to Hevo. They are not logged and never leave your host.

Notes

Auth and endpoints follow the official Hevo API docs. Pagination uses limit + starting_after where supported (see per-tool descriptions).