Claude Roboflow connectorWhat Claude can do in your Roboflow account.

Opens one conversation with the Roboflow agent, with its full message history.

Lists the conversations people in your workspace have held with the Roboflow agent, so Claude can find the right thread without you scrolling through them.

Opens one upload batch in a project and reports what Roboflow knows about it, which is how Claude checks what actually arrived.

Lists the upload batches in a project, the groups images sit in after upload and before anyone labels them.

Polls a background task by its ID, such as a project fork or a split rebalance, to see where it stands.

Returns status and progress, sub-job by sub-job, for an auto-labeling run.

Pulls up a single edge device by its ID, the record Roboflow keeps for one unit deployed in the field.

Shows the runtime configuration a device is running right now, the settings that decide what it does with its camera feeds.

Walks back through a device's earlier configuration revisions, newest first, so Claude can tell you what was different before a problem started.

Returns the default configuration your workspace applies to devices, the baseline a unit starts from before anyone tailors it.

Lists lifecycle events for a device and its streams, the timeline of starts, stops and state transitions Roboflow records.

Fetches the logs of a device, the raw lines Claude can scan for errors and patterns you would not spot by eye.

Returns aggregated hardware metrics and the overall health of a device, the numbers that tell you if the box itself is struggling.

Lists every device registered in the workspace, which gives Claude the inventory it needs before looking at any single unit.

Returns one stream configured on a device, by its ID, with the settings Roboflow holds for that feed.

Lists the streams configured on a device, one entry per camera feed or video source the unit handles.

Checks where an image zip upload stands, so Claude can tell you whether the images are in the project yet or still on their way.

Searches for images inside one project, which is how Claude finds the examples you want to look at or relabel.

Searches images across the entire workspace with RoboQL, Roboflow's query language, instead of one project at a time.

Returns the top-level summary of one model evaluation, the headline figures Claude reads before digging into details.

Returns precision, recall and F1 across confidence thresholds, the curve you need to choose where the model should draw the line.

Returns the confusion matrix of an evaluation, showing which classes the model mixes up and how often.

Returns prediction statistics image by image, in pages, so Claude can point at the pictures behind a weak score.

Returns mAP results for each split of the dataset, the standard detection score broken down by training, validation and test.

Returns performance metrics class by class for one split, which is where weak classes stop hiding behind a good average.

Returns the recommendations Roboflow generated for an evaluation, when there are any, as a starting point for the next iteration.

Returns how the images of an evaluation cluster once turned into embeddings, which helps spot groups of look-alike pictures.

Lists the model evaluations in the workspace, the index Claude uses before opening any single evaluation.

Shows the record of a trained model, with the details Roboflow keeps on how and on what it was trained.

Runs hosted inference on an image with a trained model and returns what the model sees.

Lists the trained models in a project, so Claude can compare candidates before you pick one to deploy.

Returns a project's details, including versions, classes, splits and trained models, in one view Claude can summarize.

Returns the dataset health check of a project, the view that exposes imbalanced classes or thin data.

Lists the projects in the workspace, the entry point Claude uses when you name a project loosely.

Searches images inside a public Universe dataset from its URL, so you can check what it holds before relying on it.

Searches Roboflow Universe, the public catalogue of shared datasets and models, for something close to your use case.

Returns a version's information, including its splits and the trainings that used it, which ties data to models.

Returns the full schema of one Workflow block, the inputs and outputs Claude must respect when proposing a pipeline.

Lists every available Workflow block with a short summary of each, the catalogue of building pieces for a pipeline.

Checks a Workflow JSON definition without running it, which catches mistakes before anything is saved or executed.

Opens a saved Workflow and returns the details Roboflow keeps on it, so Claude can walk you through the pipeline in plain words first.

Lists the saved Workflows in the workspace, so Claude can see which pipelines already exist before building another.