概览
What is groundlight-mcp-server?
The groundlight-mcp-server is a Model Context Protocol server for interacting with Groundlight. It provides tools to create, list, and customize Detectors, submit and list ImageQueries, create, list, and delete Alerts, and examine detector evaluation metrics. It is designed for developers building computer vision applications that can be driven by natural-language queries.
How to use groundlight-mcp-server?
Configure it by setting the GROUNDLIGHT_API_TOKEN environment variable and running the server via Docker. Add the supplied configuration block to your claude_desktop_config.json (for Claude Desktop) or settings.json (for Zed). Then invoke MCP tools such as create_detector, submit_image_query, list_alerts, and get_detector_evaluation_metrics.
Key features of groundlight-mcp-server
- Create binary, multiclass, and counting detectors.
- Submit images for analysis with confidence scores.
- Manage alerts with webhook, email, or text actions.
- Label image queries to train or correct detectors.
- View detector evaluation metrics and confusion matrices.
- Update confidence thresholds and escalation types.
Use cases of groundlight-mcp-server
- Build a computer vision application to answer natural-language questions about images.
- Automate quality inspection with alerts triggered by detector conditions.
- Continuously improve detector accuracy through feedback and manual labeling.
- Monitor video streams or image feeds with human review escalation when confidence is low.
- Integrate visual understanding into AI assistants like Claude or Zed.
FAQ from groundlight-mcp-server
What tools does the groundlight-mcp-server provide?
The server offers 14 tools: create_detector, get_detector, list_detectors, submit_image_query, get_image_query, list_image_queries, get_image, create_alert, list_alerts, delete_alert, add_label, get_detector_evaluation_metrics, update_detector_confidence_threshold, and update_detector_escalation_type.
How do I configure the server with Claude Desktop?
Add a "groundlight" entry under mcpServers in your claude_desktop_config.json using the provided Docker command and set the GROUNDLIGHT_API_TOKEN environment variable to your API token.
How can I provide labels for image queries?
Use the add_label tool with an image_query_id and a label (integer or string). For counting detectors you can optionally specify regions of interest.
What authentication is required?
The server requires a GROUNDLIGHT_API_TOKEN environment variable, passed to the Docker container at runtime.
Are there any limitations with listing image queries?
The list_image_queries tool may return a large number of results depending on the detector’s history.
