Integrations
SkillsRadar MCP: discover Skills inside your agent
Connect SkillsRadar as an MCP server so Cursor, Claude Desktop, and other hosts can search, recommend, and explain how to install verified Skills — without leaving the chat.
The short answer
- 1MCP is for discovery in chat. Skills are still installed locally with the CLI — MCP finds the right Skill and gives you the install command.
- 2Install the server (
pip install skillsradar-mcpor from the repo), then add it to your host’s MCP settings withSKILLSRADAR_API_URLand an API key. - 3Ask your agent things like “find a Skill for SaaS launch” — it should call
skillsradar_searchorskillsradar_recommend, thenskillsradar_install_help.
SkillsRadar MCP is a discovery gateway into the catalog. After you pick a Skill, install it with the agent guides or the CLI — same folders and slash commands as always.
MCP vs installing a Skill
They solve different steps of the same journey. Use MCP to find what to install; use the CLI (or install prompt) to put the Skill on disk.
Discovery
SkillsRadar MCP
- Runs as a tool server inside your agent host
- Searches and recommends from the live catalog
- Returns install commands — does not execute Skills for you
- Best when you are still deciding what to install
Execution
Installed Skill
- A folder with SKILL.md on your machine
- Loaded by Claude Code, Codex, Cursor, and others
- Works offline after install
- Best for repeatable work you already chose
Install and connect the MCP server
You need Python ≥ 3.12, a SkillsRadar account API key, and an MCP-capable host. Production API is https://api.skillsradar.dev.
Add SkillsRadar under Cursor Settings → MCP
- 1
Install the server
From a terminal with Python 3.12+:
Terminalpip install skillsradar-mcp # or, from a local checkout: # pip install -e ./apps/mcpYou should see: The
skillsradar-mcpcommand is on your PATH (or under your venv’s bin/). - 2
Create an API key
Sign in on SkillsRadar, open Account → API keys, create a key, and copy it once. Store it in env — never paste it into chat.
- 3
Add the MCP server in Cursor
Open Cursor Settings → MCP, add a server named
skillsradar, and pointcommandat the absolute path ofskillsradar-mcp:Cursor MCP JSON{ "mcpServers": { "skillsradar": { "command": "/absolute/path/to/bin/skillsradar-mcp", "args": [], "env": { "SKILLSRADAR_API_URL": "https://api.skillsradar.dev", "SKILLSRADAR_API_KEY": "sr_live_…", "SKILLSRADAR_MCP_CLIENT_NAME": "cursor" } } } }You should see: Cursor shows skillsradar as connected, with tools like skillsradar_search available.
- 4
Local development instead of production
If you run the SkillsRadar API on your machine, set
SKILLSRADAR_API_URLtohttp://localhost:5173and use a local demo or account key.
Tools your agent can call
Prefer search or recommend first, then get_skill for detail, then install_help when you are ready to put a Skill on disk.
| Tool | What it does | Typical ask |
|---|---|---|
skillsradar_search | Natural-language search over Skills | “Search SkillsRadar for SEO article skills” |
skillsradar_recommend | Recommend a workflow + Skills for a task | “Recommend Skills for a SaaS launch plan” |
skillsradar_get_skill | Full detail for one slug | “Show details for slug X” |
skillsradar_search_workflows | List or filter workflows | “List workflows about content marketing” |
skillsradar_install_help | CLI install commands + install.md link | “How do I install this for Claude Code?” |
skillsradar_trending | Trending Skills from the catalog | “What’s trending on SkillsRadar?” |
Resources: skillsradar://trending, skillsradar://skill/{slug}. Prompts: skillsradar_discover, skillsradar_recommend_task.
Environment variables
Set these in the host’s MCP env block (not in chat).
| Variable | Default | Purpose |
|---|---|---|
SKILLSRADAR_API_URL | http://localhost:5173 | API base URL — use https://api.skillsradar.dev in production |
SKILLSRADAR_API_KEY | (empty) | Developer API key for metering and recommend |
SKILLSRADAR_MCP_CLIENT_NAME | skillsradar-mcp | Display name for client telemetry |
SKILLSRADAR_MCP_CLIENT_ID | mcp-<hostname> | Stable id written to mcp_clients |
A good first conversation
Once the server is connected, stay in the agent chat. You should not need to open the website for discovery.
- 1
Describe the job, not the Skill name
Ask for an outcome. Example: “I want a repeatable workflow to research competitors and draft a launch brief.”
- 2
Let the agent call recommend or search
A well-configured host should invoke
skillsradar_recommendorskillsradar_searchand return ranked Skills / workflows. - 3
Pick a slug and ask for install help
Say which agent you use (Claude Code, Cursor, Codex). The agent should call
skillsradar_install_helpand show the CLI commands.Example installnpx --yes @skillsradar/cli@latest install <slug> --agent cursor - 4
Install locally, then run the Skill
Approve device login if needed, confirm the folder path, then invoke the Skill in your agent (
/skill-name,$skill-name, etc.). MCP does not replace that install step.
MCP not working? Match the symptom
Server shows as disconnected
Use an absolute path to skillsradar-mcp, confirm Python ≥ 3.12, and restart the host. Run skillsradar-mcp once in a terminal to surface import errors.
Tools appear but every call fails
Check SKILLSRADAR_API_URL (production vs localhost) and network access. A wrong base URL or stopped local API returns empty or HTTP errors.
Recommend fails / quota errors
Add a valid SKILLSRADAR_API_KEY from Account → API keys. Free plans have lower request limits; upgrade the developer plan if you hit caps.
Agent finds Skills but never installs them
That is expected. Ask for skillsradar_install_help, then run the CLI (or paste the install prompt). MCP discovers; the CLI installs.
Confused with “MCP Skills” elsewhere
SkillsRadar MCP exposes catalog tools. An installed Skill is still a SKILL.md folder. Connecting MCP does not auto-install Skills into ~/.agents/skills/ or ~/.claude/skills/.
FAQ
Do I need MCP if I only install Skills from the website?
No. Browse and install from the library or CLI is enough. MCP helps when you want discovery inside Cursor or Claude Desktop without switching to the browser.
Does MCP run the Skill for me?
No. It searches the catalog and explains install. After install, your agent runs the Skill the same way as any other local Skill.
Is the Developer API the same as MCP?
MCP is a thin stdio client over the same HTTP API. Scripts and products can call the API directly; agents usually prefer MCP tools.
Where do I manage API keys?
Sign in → Account → API keys. Create, rename, and revoke keys there. Never commit keys or paste them into prompts.
Also on Learn
Back to the Learn hub.
Next step
Ready to try a verified Skill?
Browse Skills compatible with this agent, or compare how folders and commands differ across Claude Code, Codex, and Cursor.