Roboflow MCP Server + Skills

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.
Get startedView Docs
Compatible with
Claude Code
Claude Code
Claude Code
Claude Code
Watch the demo
Install

One server. Every agent.

One hosted MCP server. Drop it into the agent of your choice.
Claude Connector
Hosted connector
Claude Code
CLI integration
Codex
Local agent
Cursor
IDE integration
Add Roboflow as a connector in your Claude account. Once connected, it works everywhere — Claude.ai, Desktop, and Claude Code.
  1. Go to claude.ai and click Customize in the sidebar.
  2. Click Connectors, then +, then Add custom connector.
  3. Name it , paste https://mcp.roboflow.com/mcp, and click.
  4. That’s it!
Can't run the installer? Install manually
{
  "mcpServers": {
    "roboflow": {
      "type": "http",
      "url": "https://mcp.roboflow.com/mcp",
      "headers": {
        "x-api-key": "YOUR_ROBOFLOW_API_KEY",
        "Accept": "application/json, text/event-stream"
      }
    }
  }
}
Click below to add the connector, then sign in to Roboflow when prompted
Add to Claude
Run this command in your terminal (replace YOUR_ROBOFLOW_API_KEY with your actual key):
claude mcp add -s user roboflow \
  --transport http https://mcp.roboflow.com/mcp
Can't run the installer? Install manually
{
  "mcpServers": {
    "roboflow": {
      "type": "http",
      "url": "https://mcp.roboflow.com/mcp",
      "headers": {
        "x-api-key": "YOUR_ROBOFLOW_API_KEY",
        "Accept": "application/json, text/event-stream"
      }
    }
  }
}
Add this to ~/.codex/config.toml:
[mcp_servers.roboflow]
url = "https://mcp.roboflow.com/mcp"
Can't run the installer? Install manually
{
  "mcpServers": {
    "roboflow": {
      "type": "http",
      "url": "https://mcp.roboflow.com/mcp",
      "headers": {
        "x-api-key": "YOUR_ROBOFLOW_API_KEY",
        "Accept": "application/json, text/event-stream"
      }
    }
  }
}
Add this to Cursor’s MCP config (~/.cursor/mcp.json):
{
  "mcpServers": {
    "roboflow": {
      "type": "http",
      "url": "https://mcp.roboflow.com/mcp"
    }
  }
}
Can't run the installer? Install manually
{
  "mcpServers": {
    "roboflow": {
      "type": "http",
      "url": "https://mcp.roboflow.com/mcp",
      "headers": {
        "x-api-key": "YOUR_ROBOFLOW_API_KEY",
        "Accept": "application/json, text/event-stream"
      }
    }
  }
}
⌥ Open source: roboflow/computer-vision-skills
GitHub
Tools

67 tools across 12 categories

Once connected, your agent can call any of these directly. Each tool maps to a primitive in the Roboflow platform: the same ones you'd use in the dashboard or API.
01

Agent

Chat with the Roboflow agent and build workflows
agent_chat
agent_conversation_get
agent_conversations_list
agent_workflow_publish
02

Projects

Manage projects in your workspace
projects_list
projects_create
projects_get
projects_fork
projects_health
create_project_app
03

Images

Upload and search project images
images_prepare_upload
images_prepare_upload_zip
images_upload_zip_status
images_search
04

Annotation

Save annotations and run autolabeling
annotations_save
autolabel_start
autolabel_job_get
05

Batch

Create and manage labeling jobs
annotation_batches_list
annotation_batches_get
annotation_jobs_create
06

Versions

Create and export dataset versions
versions_generate
versions_get
versions_export
07

Models

Train models and monitor training progress
models_list
models_get
models_infer
models_train
models_get_training_status
models_star_nas
trainings_get_results
trainings_stop
trainings_cancel
08

Model Evaluations

View mAP, confusion matrices, and per-class performance
model_evals_list
model_evals_get
model_evals_get_map_results
model_evals_get_confusion_matrix
model_evals_get_confidence_sweep
model_evals_get_performance_by_class
model_evals_get_image_predictions
model_evals_get_vector_analysis
model_evals_get_recommendations
09

Workflows

Build and run inference pipelines
workflows_list
workflows_get
workflows_create
workflows_update
workflow_blocks_list
workflow_blocks_get_schema
workflow_specs_run
workflows_run
workflow_specs_validate
10

Devices & Streams

Manage edge devices and video streams
devices_list
devices_get
devices_create
devices_get_config
devices_get_default_config
devices_get_config_history
devices_update_config
devices_get_logs
devices_get_events
devices_get_telemetry
devices_streams_list
devices_streams_get
11

Universe

Search public datasets on Roboflow Universe
universe_search
universe_search_app
universe_dataset_images_search
12

Misc

Poll long-running tasks and send feedback
async_tasks_get
meta_feedback_send
Not using MCP? Roboflow also ships a full CLI and REST API.
Skills

Expert knowledge, exposed to your agent.

Skills are markdown playbooks your agent reads as MCP resources. They cover auth, dataset organization, deployment choices, training playbooks, and where things live in the Roboflow app. Connected clients pick the right Skill for the task at hand.
HOW SKILLS WORK

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 its 67 tools. Prefer it offline? Install the bundle locally:
$
npx @roboflow/skills install
01

api-reference

Reference for the Roboflow REST and Inference APIs, including routes, hosts, auth, and request/response formats.
inference
rest-api
02

data-management

Use when uploading images, labeling, organizing datasets, creating Roboflow projects (detection / segmentation / keypoint / classification), tags, splits, versions, or RoboQL search.
labeling
03

inference

Use when running Roboflow model inference or choosing deployment (serverless, dedicated, self-hosted, batch). Prefers Workflows over direct model calls.
workflow-templates
workflows
04

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.
plans
pricing
05

product-navigation

Use when explaining where Roboflow features live in the app.roboflow.com web app — maps intents like upload, annotate, train, deploy to specific page URLs.
features-by-page
06

training-and-evaluation

Use when training Roboflow models or improving accuracy — covers architecture selection, model IDs, checkpoints, evaluation metrics, and the iterative improvement playbook.
improvement-playbook
07

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.
datasets
models
For AI agents

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.
View llms.txt
curl
$
curl https://mcp.roboflow.com/llms.txt
#
Roboflow MCP Server
#
Endpoint: https://mcp.roboflow.com/mcp
#
Transport: streamable HTTP
#
Auth: x-api-key header
#
67 tools across 12 categories…
Ready to give your agent vision?
Get startedView Docs

Additional resources

Vision AI is transforming every industry. Let’s transform yours.

Roboflow MCP: Build Vision Apps with Claude, Codex, and More

The new Roboflow MCP Server connects Claude, Codex, and other AI agents to Roboflow so they can label visual data, train custom models, and build deployable vision pipelines.
FAQ

Frequently asked questions.

Quick answers about the server, Skills, auth, and how MCP fits into the rest of Roboflow. Don't see yours? Check the docs.
What is the Roboflow MCP Server?

The Roboflow MCP Server is a hosted Model Context Protocol (MCP) server at mcp.roboflow.com that gives AI coding agents direct access to a Roboflow workspace. It exposes 67 tools across projects, datasets, training, Workflows, and Universe, so any MCP-compatible client (Claude, Claude Code, Cursor, Codex, Windsurf, Continue) can manage computer vision projects through natural-language prompts. Connect by pasting the config snippet for your agent (Claude Code, Claude Desktop, Codex, or Cursor) from the install section above. Skills are expert markdown playbooks the agent reads as MCP resources, and they install separately with npx @roboflow/skills install.

‍

What's the difference between MCP and Skills?

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, data-management, inference, plans-and-pricing, product-navigation, training-and-evaluation, and universe.

How is this different from the Roboflow REST API?

The REST API is for code. MCP is for agents. The server speaks streamable HTTP at /mcp and presents 67 tools your AI agent can discover and call directly. It is tuned for natural language to tool call workflows rather than hand-written integrations.

Which agents are supported?

Install snippets are provided for Claude Code, Claude Desktop, Codex, and Cursor. The Skills CLI installs standalone Skills into Claude Code, Cursor, OpenCode, and any agent that reads SKILL.md files. The MCP server itself works with anything that speaks MCP over streamable HTTP.

Do I need to install anything locally?

The MCP server is hosted at mcp.roboflow.com, so it runs in the cloud. To connect, paste the config snippet for your agent (or run the claude mcp add command for Claude Code) from the install section above. To also use Skills locally, install the bundle with npx @roboflow/skills install.

How do I authenticate?

Paste your workspace API key into the x-api-key header of the config snippet for your agent (or pass it via the --header flag for the Claude Code CLI command). Grab a key from app.roboflow.com/settings/api.

Can I install per project?

Yes. For Claude Code, swap -s user for -s project in the claude mcp add command to scope the server to the current project. For other agents, place the config snippet in the project-local config file your client supports (e.g., Cursor reads .cursor/mcp.json in the project root).

Is it free?

The MCP server and Skills are free and open source under Apache-2.0. You only pay for the underlying Roboflow usage on the same plans you already have.

Where can I report a bug or suggest a Skill?

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.