Provides access to the Roboflow API to manage computer vision projects, including tools for listing workspaces and projects, uploading images with annotations, generating and downloading dataset versions, searching Roboflow Universe for public data, and running inference on trained models.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@Roboflow MCP Server search Universe for public datasets of solar panels"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
A Model Context Protocol (MCP) server that exposes the Roboflow platform API as tools in Claude Code. Manage datasets, trigger training runs, search Universe, and run inference — all from the CLI.
Requirements: Python 3.10+, a Roboflow API key, Claude Code installed.
git clone https://github.com/nickedridge-wq/roboflow-mcp.git
cd roboflow-mcp
python -m venv venv
source venv/bin/activate
pip install -r requirements.txtOption A — project-level (recommended, checked into the repo):
claude mcp add roboflow \
--env ROBOFLOW_API_KEY=your_api_key_here \
-- /path/to/roboflow-mcp/venv/bin/python /path/to/roboflow-mcp/server.pyThis writes a .mcp.json file in the current project directory.
Option B — user-level (available in all projects):
claude mcp add roboflow --scope user \
--env ROBOFLOW_API_KEY=your_api_key_here \
-- /path/to/roboflow-mcp/venv/bin/python /path/to/roboflow-mcp/server.pyRestart Claude Code — the mcp__roboflow__* tools will be available immediately.
Related MCP server: XLMCP
Tool | Description |
| Show workspace name, URL slug, and project count |
| List all projects in a workspace with type and image counts |
| Get classes, annotation type, and metadata for a project |
| List all dataset versions with image counts per split |
| Upload an image and optional annotation to a project |
| Generate a new dataset version with preprocessing and augmentation |
| Download a version locally (yolov8, coco, voc, and more) |
| Download a public dataset directly from Roboflow Universe |
| Search Universe for public datasets and pre-trained models |
| Run inference via a deployed model on a local file or URL |
| Fetch mAP, precision, and recall for a trained version |
search_universe("hard hat detection")
→ pick a result, note workspace + project + version
download_universe_dataset(
universe_workspace="roboflow-universe-projects",
universe_project="hard-hat-universe",
version_number=1,
model_format="yolov8",
location="./datasets/hard-hat"
)upload_image(project_url="my-project", image_path="/data/img001.jpg",
annotation_path="/data/img001.xml", split="train")
create_version(
project_url="my-project",
preprocessing={"auto-orient": True, "resize": {"width": 640, "height": 640, "format": "Stretch to"}},
augmentation={"flip": {"horizontal": True}, "rotation": {"degrees": 15}}
)run_inference(project_url="my-project", version_number=3,
image_path="/data/test.jpg", confidence=60)
get_model_metrics(project_url="my-project", version_number=3)python -m unittest test_server -v21 tests covering output suppression, lazy init thread safety, input validation, null model guard, auth error propagation, and parameter contracts. No live API key required.
Lazy authentication — the Roboflow SDK authenticates once per session on first tool call, with double-checked locking for thread safety.
Output suppression — the SDK prints to both stdout and stderr on init, which corrupts MCP's stdio transport. All SDK calls redirect both streams.
search_universe and get_model_metrics call the Roboflow REST API directly for endpoints not exposed cleanly through the SDK.
This server cannot be deployed
Roboflow computer vision for AI agents: datasets, annotation, versioning, workflows, inference.
Source-checked CLI guides and model-aware planning for Claude Code, Codex, and Grok Build.
Deploy, monitor, and manage your OpenClaw AI assistants via natural language.
Connect Claude, Cursor, Codex, and other AI tools to your robotics mission data.
