Computer vision for your coding agent.
Give your coding agent direct access to Roboflow. Train custom models, run inference, and manage datasets, all from a prompt.
Install
One server. Every agent.
One hosted MCP server. Drop it into the agent of your choice.
Paste it into your coding agent and it walks you through the rest. Setting up by hand? The server is https://mcp.roboflow.com/mcp.
Open the Roboflow listing, select Connect, then sign in to Roboflow. Available on desktop, web and mobile.
Run these commands in your terminal, then sign in to Roboflow. Claude Code CLI has no documented browser install link.
Opens the Roboflow plugin listing. Sign in if prompted and confirm the install on Cursor to add MCP and skills.
Select Add > Create custom MCP server. Name it Roboflow, paste the server URL, select OAuth, and sign in to Roboflow. Developer mode may be required; availability depends on your account and workspace settings.
Open Plugins in Codex, choose the Roboflow source, and install the Roboflow plugin.
Then run /mcp auth roboflow in Gemini CLI to sign in. Gemini CLI has no documented browser install link.
231 tools for every step, from dataset to deployment.
Once connected, your agent calls these directly. Each tool maps to something you’d otherwise do in the dashboard or the API. The tool reference lists every one.
| Category | What your agent can do | Tools |
|---|---|---|
| Agent | chat with the Roboflow AI agent for Roboflow Q&A, advanced Workflow building, and CV solution planning
| 5 |
| Projects | manage projects in your workspace
1 moreFewer
| 7 |
| Project Deployment | launch and manage a stable live endpoint for a project
2 moreFewer
| 8 |
| Images | upload, search, and manage images across projects
1 moreFewer
| 7 |
| Media Understanding | analyze images and short videos with a vision model to scope CV tasks, suggest classes, and plan workflows
| 4 |
| Annotations | save image annotations and Action Recognition video segments
| 2 |
| Auto-label | kick off and track hosted auto-labeling jobs
| 4 |
| Annotation Administration | explicitly manage annotation batches, assignments, review, and Dataset acceptance
17 moreFewer
| 23 |
| Versions | create and inspect dataset versions
| 4 |
| Models | train models and monitor training progress
6 moreFewer
| 12 |
| Model Evaluations | inspect held-out evaluation results for trained models — mAP, precision/recall sweeps, per-class breakdowns, confusion matrices, vector clustering, per-image predictions, and accuracy/latency comparisons
4 moreFewer
| 10 |
| Workflows | build and execute inference pipelines
4 moreFewer
| 10 |
| Workflow Evals | design evaluations for Workflows: discover evaluators and schemas, author Specs and Eval Datasets, manage Cases, and confirm Workflow output bindings
31 moreFewer
| 37 |
| Workflow Eval Runs | run Workflow Evals against one or more Workflows, prepare candidate outputs, inspect and compare per-Case results, export them, and analyze image embeddings
22 moreFewer
| 28 |
| Batch Processing | run Workflows over huge image/video sets: Asset Library selections (platform-staged) or batches you stage yourself
7 moreFewer
| 13 |
| Devices | list, inspect, and provision Roboflow Deployment Manager devices via the external Deployments v2 API
13 moreFewer
| 19 |
| API Keys | manage workspace API keys
3 moreFewer
| 9 |
| Folders | list the workspace folders visible to the credential
| 1 |
| Vision Events | query and report on production vision events (quality checks, inventory counts, safety alerts, operator feedback) and manage the use cases they are grouped under
2 moreFewer
| 8 |
| Universe | search public datasets on Roboflow Universe
| 1 |
| Datasources | configure and run cloud-bucket mirroring into projects
3 moreFewer
| 9 |
| Credentials | manage cloud-storage credentials used by datasources
| 3 |
| Trash | list and restore soft-deleted items
| 5 |
| Async tasks | poll the status of long-running async operations
| 1 |
| Meta | report issues or suggestions
| 1 |
11 Skills, exposed to your agent.
Skills are markdown playbooks your agent reads as MCP resources. Connected clients pick the right one for the task at hand.
- 01
api-reference
Protocol-level facts for Roboflow REST and Inference APIs — URL patterns, auth, parameters, error codes, and SDK quick-start. For deployment strategy and Workflow execution patterns, see roboflow-inference.
- 02
batch-processing
Use when running a Roboflow Workflow over many images or videos with batch processing — Asset Library jobs (the platform stages workspace images for you) or staged jobs (you stage external local/cloud files with the inference-cli), plus monitoring jobs, downloading results, and routing imports to datasources (bucket mirror) instead.
- 03
cloud-storage
Use when connecting cloud storage (AWS S3 / S3-compatible or Google Cloud Storage) to Roboflow to mirror images into a workspace — creating storage credentials, defining datasources (bucket-mirror configs), selecting objects with glob rules, validating access, and running/scheduling mirror jobs.
- 04
custom-weights-upload
Use when uploading locally trained model weights (YOLO, RF-DETR, YOLO-NAS, PaliGemma, Florence-2) to Roboflow — the client-side Python SDK upload flow, safe API key handling, per-family packaging requirements, and verifying the upload.
- 05
data-management
Use when uploading images, labeling, organizing datasets, creating Roboflow projects (detection/segmentation/keypoint/classification), tags, splits, versions, or RoboQL search.
- 06
inference
Deployment option comparison (serverless, dedicated, self-hosted, batch) and Workflow execution patterns. For raw API URL patterns, auth, and request/response formats, see roboflow-api-reference.
- 07
plans-and-pricing
Use when answering questions about Roboflow plans, credit usage, or cost estimation; directs users to roboflow.com/pricing for current dollar amounts.
- 08
product-navigation
Use when explaining where Roboflow features live in the app.roboflow.com web app, mapping intents like upload, annotate, train, deploy to specific page URLs.
- 09
training-and-evaluation
Use when training Roboflow models, diagnosing why a model underperforms, improving accuracy, or setting up a production feedback loop — covers architecture selection, model IDs, checkpoints, Model Evaluation (confusion matrix, per-class metrics, MCP eval tools), the diagnosis-first improvement playbook, and Active Learning through the Project Model Workflow block.
- 10
universe
Use when searching for or using public datasets/models on Roboflow Universe (universe.roboflow.com), the open repository of 1M+ computer vision datasets and 50K+ pre-trained models.
- 11
workflow-from-eval
Build or improve a Roboflow Workflow using eval-driven search. Prefer this for open-ended business problems where the approach is unclear, or when comparing alternatives. A business goal plus images or video is enough to start; no existing Workflow or Eval is required. Build an example suite and scorecard the user approves, compare candidate Workflows, then tune and validate the winner. Use direct tools for straightforward builds and edits, Workflow Evals tools for measuring one existing Workflow once, and training skills for training a model.
Skills
Skills load when your agent needs them. No setup on your end.
Once your agent is connected to mcp.roboflow.com, it can list and read every Skill alongside the tools. Want them as local files instead? Install the bundle.
llms.txt
A short, LLM-friendly summary lives at /llms.txt
Quick answers about the server, Skills, auth, and how MCP fits into the rest of Roboflow.
Additional resources
Featured4 min readRoboflow MCP: Build Vision Apps with Claude, Codex, and More
Patrick DeschereMay 20, 2026
Frequently asked questions
A hosted Model Context Protocol server at mcp.roboflow.com that gives AI coding agents direct access to your Roboflow workspace. Its tools cover projects, datasets, labeling, training, evaluation, Workflows, deployment and Universe, so any MCP client can run computer vision work from a prompt. Connect with the install steps above. Skills, the expert playbooks your agent reads, come with it.
MCP gives your agent the tools (verbs). Skills give it the expert knowledge (when, why and how). Tools are live primitives like “create a project” or “run a Workflow”. Skills are durable markdown playbooks: api-reference, batch-processing, cloud-storage, custom-weights-upload, data-management, inference, plans-and-pricing, product-navigation, training-and-evaluation and universe.
The REST API is for code. MCP is for agents. The server speaks streamable HTTP at /mcp and presents its tools for your agent to discover and call directly, so a sentence becomes a tool call without a hand-written integration.
Install steps are above for Claude, Claude Code, Codex and Cursor. Gemini CLI, Devin, Cline and any other client that speaks MCP over streamable HTTP connect with the server URL: https://mcp.roboflow.com/mcp. The Skills CLI installs Skills into any agent that reads SKILL.md files.
No. The server is hosted at mcp.roboflow.com, so it runs in the cloud and your agent connects with the install steps above. To keep the Skills as local files too, run npx skills add roboflow/computer-vision-skills.
You sign in to Roboflow. The server uses OAuth, so there is no API key to copy: the first time your agent connects, it opens Roboflow in your browser to sign in and authorize it. Connecting through a gateway such as Azure AI Foundry or Copilot Studio? Register an OAuth app under Workspace Settings > Developer.
Yes. In Claude Code, add --scope local to claude plugin install roboflow. In Cursor, put the server in .cursor/mcp.json at the project root rather than ~/.cursor/mcp.json. In Gemini CLI, use --scope project on gemini mcp add.
The MCP server and Skills cost nothing extra: you pay only for the Roboflow usage behind them, on the plan you already have. The Skills are open source under Apache-2.0.
Open an issue or PR on roboflow/computer-vision-skills. Skills are markdown; contributions are mostly editing or adding files under skills/, each with a SKILL.md frontmatter block.