For agents

Give your agent the skill atlas

The Skillspace MCP server lets an AI agent search 27,503 launch-quality skills by the problem it needs to solve, and read our rating of each one before it installs anything.

https://mcp.skillspace.sh/mcp

Set it up with one prompt

Paste this into your agent. It will connect the server, read the Skillspace skill and follow it, and install nothing without asking you.

Set up Skillspace for me. It indexes public agent skills and rates each one for quality and safety.

1. Connect its MCP server at https://mcp.skillspace.sh/mcp (Streamable HTTP, no sign-in). In Claude Code, run:
   claude mcp add --transport http skillspace https://mcp.skillspace.sh/mcp
   If you can't add it yourself, tell me what to do.
2. Read https://skillspace.sh/mcp/SKILL.md and follow it whenever I need a capability you don't have, or ask for a skill, tool or workflow for a task.
3. If you can load skills, offer to save that file as skillspace/SKILL.md in your skills folder, so you have it next time.
4. Never install or run a skill until you have read its source on GitHub and I have said yes. Never install or run one rated risky or unsafe, or one with a safety flag.

Then tell me whether the Skillspace tools are available.

Connect it

Claude Code

claude mcp add --transport http skillspace https://mcp.skillspace.sh/mcp

Cursor, and other clients with a JSON config

{
  "mcpServers": {
    "skillspace": {
      "url": "https://mcp.skillspace.sh/mcp"
    }
  }
}

Anything else

It is a stateless Streamable HTTP server with no sign-in. It speaks MCP 2026-07-28, and 2025-03-26 to 2025-11-25 through initialize.

The Skillspace skill

A short Agent Skill that teaches your agent to use the server well: search by the problem, read the rating and then the source on GitHub, recommend with reasons, and install nothing without asking you.

Download the Skillspace skill

Claude Code

Save it as ~/.claude/skills/skillspace/SKILL.md:

mkdir -p ~/.claude/skills/skillspace
curl -fsSL https://skillspace.sh/mcp/SKILL.md -o ~/.claude/skills/skillspace/SKILL.md

Any other agent

Any agent that reads Agent Skills can use it. Save it as skillspace/SKILL.md in that agent's skills folder.

Tools

find_skills {problem, limit}

A plain-language problem in, up to 10 rated skills out, ranked for relevance, each with its grade, safety verdict, safety flag, the evidence behind the rating, and links. limit is 1–10, default 5. The answer says ranking (ai-reranked, text-match, or none when nothing matched) and confident, and adds a note when unsure.

get_skill {skill}

One skill's full rating and review: score breakdown, scanner findings, safety flag and exclusion reason, judge reviews and head-to-head results where they exist, licence and adoption. skill is a slug or a 16-character id.

browse_category {category, limit}

The best-rated launch-quality skills in a category. limit is 1–50, default 10.

list_categories {}

The category ids, names, teams and counts.

Safety verdicts and flags

Every result carries safety, the verdict of our static scan: ok, review, risky or unsafe. It also carries safety_flag, which is null unless the skill is flagged: prompt-injection, guardrail-bypass, harmful-content, credential-exposure, attack-benchmark or test-fixture. The first four are what our model flagged the skill for; an attack-benchmark is a set of test cases written to attack AI agents, and a test-fixture a test file not meant for use, some of them deliberately malicious. A flag overrides an ok verdict. A non-null safety_flag is a stop, like a risky or unsafe verdict: the Skillspace skill tells your agent not to install, run or recommend the skill, and to tell you what was flagged.

Limits

20 requests a minute and 1,000 a day per client. A client is one IPv4 address or one IPv6 /48 network, so people behind one shared address, such as a company network or a hosted agent platform, share one client's limits. The whole server answers at most 50,000 requests a day; past that, every request gets a 429 until midnight UTC. AI-assisted ranking has a shared daily budget; past it, results are ranked by text match rather than refused. No tool returns more than 50 skills.

Privacy

Problem text is logged anonymously (never with an IP or account) after emails, secret keys, passwords written as a setting or said with a digit or symbol, card numbers and phone numbers in common formats are scrubbed. A setting is password=, pass=, pw=, a key such as DB_PASSWORD, or a JSON "password" key; said means "my password is" and a quoted phrase or a word with a digit or symbol. A password of letters alone, or the rest of an unquoted phrase, is not caught, so don't put personal data or secrets in it. What is kept: the scrubbed text, at most its first 256 characters, in our search-demand analytics, with the category of the top result and how the results were ranked; and its first 300 characters, scrubbed the same way, in the log of the Cloudflare AI Gateway through which the ranking model is called. The server keeps no request logs, and its error logs say what failed, never what was searched. The rate limits count each client under a one-way hash of its IP address, or of its IPv6 /48 network, mixed with a secret and the date; those counts are deleted the next day. The privacy page has the details.

What the ratings are

Our opinion, produced by the published scoring engine and AI judging. Every rating describes the version of a skill that Skillspace crawled for its 24 September 2026 snapshot. The repository may have changed since, so go by its current files, not by the rating. Skillspace indexes and rates skills; it does not host or endorse them. Always read the source on GitHub before installing anything. The terms cover the rest.