Roboflow MCP server

Connect Roboflow to Claude, ChatGPT, Cursor, Claude Code, Codex, Gemini, GitHub Copilot and any other MCP client, for your whole team. Metorial adds security, governance, and tracing, and gives everyone one Magic MCP URL to connect.

What you can do with the Roboflow MCP server

Ask your agent in plain language. Each of these maps to a real tool on the server.

  • Use the list annotation jobs tool in Roboflow
  • Use the upload annotation tool in Roboflow
  • Use the run inference tool in Roboflow
  • Use the create annotation job tool in Roboflow
  • Use the get project tool in Roboflow
  • Use the train model tool in Roboflow

Supported Tools

list_annotation_jobs

List Annotation Jobs

List all annotation jobs for a project. Returns each job's status, image counts, and progress towards completion. Optionally retrieve detailed information for a specific job by providing a jobId.

upload_annotation

Upload Annotation

Upload an annotation file for a specific image in a project. Supports various annotation formats including YOLO/Darknet, Pascal VOC XML, COCO JSON, and more. The annotation text should match the expected format for the project type.

run_inference

Run Inference

Run predictions on an image using a trained model deployed on the Roboflow hosted inference API. Provide a publicly accessible image URL or base64-encoded image data. Returns detected objects with bounding boxes, confidence scores, and class labels.

create_annotation_job

Create Annotation Job

Create a new annotation job for a project. An annotation job assigns images from a batch to a labeler for annotation and a reviewer for quality control. Both must be members of the workspace.

get_project

Get Project

Get detailed information about a specific Roboflow project, including its dataset versions, annotation classes, image splits, and training status. Use the project URL slug as the projectId.

train_model

Train Model

Start training a model on a specific dataset version using Roboflow Train. Training runs asynchronously and typically completes within 24 hours. Once done, the model is automatically deployed to a hosted inference API. Use **Get Version** to monitor training progress and results, and **Stop Training** to end it early.

list_projects

List Projects

List all projects in a Roboflow workspace. Returns project metadata including name, type, image counts, annotation classes, and version counts. Useful for discovering available projects and their current state.

export_dataset

Export Dataset

Export a dataset version as a downloadable ZIP in a specified annotation format. Supported formats include YOLOv5, COCO JSON, Pascal VOC, TFRecord, and many more.

get_image

Get Image

Get detailed information about a specific image in a project, including its annotations, labels, tags, split assignment, and URLs for the original and thumbnail versions.

delete_images

Delete Images

Delete one or more images from a Roboflow project. This permanently removes the images and their annotations from the project.

create_project

Create Project

Create a new computer vision project in the workspace. Supports object detection, classification (single and multi-label), instance segmentation, and semantic segmentation project types.

create_version

Create Version

Generate a new versioned snapshot of a project's dataset with optional preprocessing and augmentation settings. The version is generated asynchronously. Use **Get Version** to check generation progress. Preprocessing options include auto-orient, resize, grayscale, static crop, and tile. Augmentation options include flip, rotate, crop, brightness, blur, noise, cutout, mosaic, and more.

search_images

Search Images

Search for images within a project using text prompts, visual similarity to an existing image, or by filtering on tags, classes, and batch assignments. Supports pagination for browsing large datasets.

upload_image

Upload Image

Upload an image to a Roboflow project. Provide either a publicly accessible image URL or base64-encoded image data. Optionally assign the image to a batch, add tags, or specify the dataset split.

get_version

Get Version

Get detailed information about a specific dataset version, including preprocessing and augmentation settings, training results (mAP, precision, recall), and model endpoint details. Use this to check training progress or model performance.

manage_image_tags

Manage Image Tags

Add, remove, or set tags on a specific image in a project. Tags are used for organizing and filtering images. Use "add" to append tags, "remove" to delete specific tags, or "set" to replace all existing tags.

list_batches

List Image Batches

List uploaded image batches for a project, including IDs and names to use with create_annotation_job. Call list_projects to discover projects.

who_am_i

Get Authenticated Workspace

Verify the connected credential and return the workspace it belongs to.

delete_project

Move Project to Trash

Move a project to Trash, hiding it from the workspace and cancelling its in-flight training jobs. Roboflow retains it for 30 days and allows restoration from Trash during that period. Call list_projects to discover projects.

stop_training

Stop Training

End a model training job on a dataset version before it finishes. Use **stop** to finish early and keep the current model weights, or **cancel** to discard the weights (credits may be refunded if canceled early). Use **Get Version** to confirm the resulting training status.

list_models

List Models

List trained models in a project, including models trained on dataset versions and standalone models such as Neural Architecture Search children. Returns each model's ID, architecture, training status, source version, and evaluation metrics. Call list_projects to discover projects.

Connect Roboflow to your AI client

Every client uses the same Magic MCP URL, and so does any other MCP client. Add it once and sign in with SSO.

  1. Enable Roboflow in Metorial. In your Metorial portal, enable the Roboflow MCP server and sign in with the account that owns the data. Access stays governed by your team policies.
  2. Add a custom connector. Open Customize, then Connectors. Select Add, then Custom, then Web.
  3. Paste your Metorial MCP URL. Name it Metorial, paste your Magic MCP URL (<your-magic-mcp-url>) and continue.
  4. Sign in. Select Add and sign in to load Roboflow tools.

Run Roboflow through the Metorial gateway

One governed layer for every app your agents reach, not a separate connector per tool.

  • Give Roboflow and every other app to your whole team through one MCP gateway, behind your SSO.
  • Scope which teams and agents can use Roboflow tools with role-based access control, and trace every call.
  • Prefer to run it yourself? Metorial is open source and self-hostable.

More integrations teams use with Roboflow

Google Sheets

Create, read, and update spreadsheets and their cell data. Read and write values to individual cells, ranges, or multiple ranges using A1 notation or named ranges. Apply cell formatting including text styles, backgrounds, borders, number formats, and conditional formatting. Manage sheets (tabs) within a spreadsheet—add, delete, copy, rename, and reorder them. Create and manage embedded charts, pivot tables, filter views, data validation rules, protected ranges, named ranges, and merged cells. Perform batch operations to apply multiple updates atomically.

GitHub

Manage repositories, issues, and pull requests. Create and configure branches, star repositories, review code, and merge changes. Automate CI/CD workflows with GitHub Actions, manage workflow runs, secrets, and artifacts. Track issues with labels, milestones, and assignees. Search across code, repositories, issues, and users. Manage organizations, teams, and memberships. Create and manage projects, gists, packages, deployments, and environments. Access security alerts including code scanning, secret scanning, and Dependabot alerts. Read and write file contents in repositories. Manage webhooks, notifications, and codespaces.

Salesforce

Manage CRM data including Accounts, Contacts, Leads, Opportunities, Cases, and custom objects. Create, read, update, and delete records. Query data using SOQL and search across objects using SOSL. Perform bulk data operations for large-scale imports, exports, and migrations. Execute composite requests to batch multiple operations in a single API call. Access analytics, reports, and dashboards. Manage files and attachments associated with records. Interact with Chatter feeds, posts, and groups for social collaboration. Manage org metadata including custom objects, fields, layouts, and workflows. Query data using GraphQL for precise data retrieval across related objects.

Sharepoint

Manage SharePoint sites, document libraries, lists, and files. Create, read, update, and delete lists and list items with custom columns. Resolve site users to numeric Person/Group LookupId values. Upload, download, move, copy, and version files in document libraries. Search across sites, files, folders, lists, and list items using Microsoft Search. Manage permissions at site, list, and item levels with granular access control. Define and manage content types and site columns. Retrieve site properties and search for sites across Microsoft 365.

Airtable

Create, read, update, and delete records in Airtable bases and tables. Manage base schemas including creating tables and fields. Filter records using formulas, sort by fields, and scope queries to specific views. Upsert records to find, create, or update in a single call. Upload attachments to records, read and write record comments, list accessible bases, and receive real-time base change events through webhooks.

Google Drive

Upload, download, create, copy, move, rename, trash, and permanently delete files and folders in Google Drive. Search for files using complex queries filtering by name, MIME type, owner, modification date, labels, and other metadata. Share files and folders with specific users, groups, or domains with role-based permissions (owner, writer, commenter, reader). Manage shared drives and their members. Export Google Workspace files (Docs, Sheets, Slides) to standard formats like PDF, DOCX, and XLSX. Track file revision history and restore earlier versions. Create, read, update, and delete threaded comments and replies on files. Read label definitions and search files by label. Browse the Drive apps installed for your account. Store app-specific data in a hidden per-user folder.

Technical notes for Roboflow

Manage Roboflow computer vision projects and image datasets. Discover projects and image batches, upload and annotate images, search and tag datasets, generate and export dataset versions, start model training, run hosted object detection inference, and manage annotation jobs.

Frequently asked questions

Common questions about connecting Roboflow to AI agents with Metorial.

  1. How many tools does the Roboflow MCP server have?
    The Roboflow MCP server exposes 21 tools, including List Annotation Jobs, Upload Annotation, Run Inference. Each one is governed and traced through Metorial.
  2. Roboflow exposes tools your agents call on demand; it does not publish event triggers yet. Everything still runs through one governed MCP URL.
  3. Any MCP client, including VS Code, Windsurf, Zed, OpenCode and Grok. Each one reaches the Roboflow MCP server through the same Metorial URL.
  4. Use role-based access control to decide which teams, users, and agents can use Roboflow and which of its tools they can call.
  5. Yes. Metorial is open source, so you can run the Roboflow MCP server yourself or use the hosted gateway.

Connect Roboflow to production AI agents

See how Metorial gives your team governed access to Roboflow, with tracing and security controls built in.