SkillDB
Model Context Protocol

One install. Your agent already knows.

The SkillDB MCP server connects 5,900+ expert skills to any MCP-compatible agent. Add it once. Search, install, and apply specialist knowledge from inside your tool of choice — no manual prompt engineering, no copy-paste, no context juggling.

MCP CONNECTOR URL
https://mcp.skilldb.dev

Works with

Setup

Three steps. Sixty seconds.

Pick your agent. We'll show you exactly what to do.

Settings → Connectors → Add custom connector

1

Open Claude Desktop

In Claude Desktop, open Settings, then Connectors. Click "Add custom connector" at the bottom of the list.

2

Paste the SkillDB connector URL

https://mcp.skilldb.dev
3

You're done. Try it.

Click Connect, sign in with your SkillDB account, and approve the requested tools. SkillDB now appears as an active connector — your agent can search, install, and use any of 5,900+ skills.

Full walkthrough with screenshots

Tools your agent gets

10 tools, 5,900+ skills

One install registers all of these. Your agent invokes them on demand.

skilldb_search

Semantic search across 5,900+ skills with relevance ranking.

skilldb_get

Fetch the full markdown of a specific skill by id or path.

skilldb_list

Browse skills filtered by category, pack, or tag.

skilldb_recommend

Suggest skills given the current task or technology stack.

skilldb_suggest

Propose new skills the user might want to author.

skilldb_my_skills

List skills in the user's private Studio workspace.

skilldb_create_skill

Author a new private skill from the agent.

skilldb_update_skill

Update an existing private skill.

skilldb_purge

Clear cached skill content; force a fresh fetch.

skilldb_set_key

Configure or rotate the API key mid-session — no restart.

What it's for

Your agent, the specialist

Generic completion becomes expert-grade work. Same model, different output.

Code review with teeth

Load the software-skills pack and security-practices-skills. Every PR gets a senior-engineer review with concrete actionable feedback.

Incident response at 3 AM

incident-postmortem-skills + security-monitoring covers severity classification, runbooks, blameless writeups. Your agent runs the timeline reconstruction while you handle stakeholders.

Specialist creative work

author-archetypes, screenplay-format-skills, copywriting-skills — your agent writes in a register that doesn't collapse to generic LLM voice.

Domain-specific compliance

Pull regulatory-compliance-skills, gdpr-skills, hipaa-skills exactly when needed. The agent knows which framework applies to which question.

On-call without burnout

incident-response-runbooks + customer-communication. The agent drafts the status page update while you investigate root cause.

Greenfield architecture

system-design-skills, threat-modeling-in-design, microservices-decomposition. Architecture review before the first line of code.

FAQ

Common questions

What does SkillDB add to my agent?
A registry of 5,900+ peer-authored skills across 38 domains — code review, security audits, data engineering, design systems, technical writing, incident response, and dozens more. Skills are fetched on demand into the agent's context, so the model has expert-grade guidance for whatever task is in front of it.
Which agents are supported?
Anything that speaks Model Context Protocol. We test against Claude Desktop, Claude Code, Cursor, OpenClaw, Hermes, and NemoClaw. Most other MCP-compatible tools work without changes — if your agent reads MCP servers, it can read SkillDB.
Do I need an API key?
You can browse skill metadata and search the catalog without one. Full skill content (the markdown that goes into the agent) requires a free API key — generate one on /api-access or run skilldb login from the CLI. Pro and Studio plans get higher rate limits.
Is the connection self-hosted or cloud?
Both options work. The default at https://mcp.skilldb.dev is the managed connector — fastest path, runs on Cloud Run, no install. The npx skilldb-mcp variant runs the same server locally over stdio and is the choice for offline-capable workflows or stricter network policies.
How does authentication work?
For Claude Desktop and other connector-style clients, OAuth — sign in with your SkillDB account on first connect. For stdio clients (Claude Code, Cursor), set the SKILLDB_API_KEY environment variable or pass --api-key on the command line.
Will SkillDB skills work in my agent's context window?
Yes. Skills are scoped — most are 200 to 500 lines of focused markdown. The MCP server returns just the skills the agent asks for, on demand, so your context stays lean.
Can I publish my own skills via MCP?
On the Studio plan, yes — the skilldb_create_skill and skilldb_update_skill tools let your agent author and update private skills directly. Public publishing goes through the web app review queue.
What's the rate limit?
Free: 50 calls/day. Pro ($9/mo): 10,000 calls/month. Studio ($29/mo): 50,000 calls/month plus private skills. Burst limits apply to all plans to protect the service.
How do I uninstall?
Claude Desktop: remove the connector from Settings → Connectors. Claude Code: claude mcp remove skilldb. Cursor: delete the entry from your MCP settings JSON. The SkillDB account remains; you can reconnect any time.

Ready to install?

Free tier covers most agents. No credit card. Sign in with Google or email and get an API key in under a minute.

Get your API keyBrowse the catalog

Stuck? Read the Claude Desktop walkthrough · Full docs