#How to install SkillDB in Claude Desktop in under 60 seconds

Claude Desktop can run MCP servers straight from its config file — so adding SkillDB takes one small block of JSON and a restart. No browser extension, no account wall, no copy-pasting skills by hand. Once it's in, your agent can search a library of 5,900+ specialist skills and pull the right one into context on demand.

This is the walkthrough. If you'd rather just do it: open Claude Desktop → Settings → Developer → Edit Config, drop the skilldb block below into mcpServers, save, and restart. Done.

#What you need before you start

• Claude Desktop, a recent version (the Developer / Edit Config option is in current builds).
• Node.js 18+ installed — the server runs via npx, which ships with Node.
• (Optional) a SkillDB API key. You can search and browse the catalog with no key at all. Full skill content uses a free key — grab one at skilldb.dev/api-access. Free tier is 50 calls/day, no card.

#Step 1: Open the config

Launch Claude Desktop. Open Settings → Developer and click Edit Config. That opens claude_desktop_config.json in your editor. (On macOS it's at ~/Library/Application Support/Claude/claude_desktop_config.json; on Windows at %APPDATA%\Claude\claude_desktop_config.json.)

If the file is empty, start with { "mcpServers": {} }.

#Step 2: Add the SkillDB server

Add a skilldb entry under mcpServers:

{
  "mcpServers": {     "skilldb": {       "command": "npx",       "args": ["-p", "skilldb", "skilldb-mcp"]     }   } }

That's the whole config. The -p skilldb tells npx to run the skilldb-mcp binary from the skilldb npm package.

Want full skill content without entering a key each time? Add an env block:

{
  "mcpServers": {     "skilldb": {       "command": "npx",       "args": ["-p", "skilldb", "skilldb-mcp"],       "env": { "SKILLDB_API_KEY": "your_key_here" }     }   } }

#Step 3: Restart and try it

Save the file, then fully quit and reopen Claude Desktop (a window reload isn't always enough). SkillDB now shows up in your tools. The first run, npx downloads the package; after that it's cached.

Start a new conversation and ask:

• "Review this Python script for me. Use SkillDB to find a relevant code-review skill first."
• "Help me write a blameless postmortem for last week's outage. Pull in a SkillDB incident-postmortem skill."
• "I'm building an MCP server in TypeScript — what SkillDB skills would help?"

You'll see the tool-call indicator fire as Claude reaches into SkillDB (skilldb_search, then skilldb_get on the best match). The augmented answer tends to be noticeably more specific and grounded than the same question asked without it.

#What just happened

The skilldb-mcp server runs locally on your machine over stdio and talks to the SkillDB API to fetch skills. Each conversation, Claude reads its tool schema and gains ten tools — skilldb_search, skilldb_get, skilldb_list, skilldb_recommend, and six more. When a task would benefit from specialist knowledge, the model searches the catalog, loads the matching skill into context, and works from it. You don't orchestrate any of that — the model decides; you just get better output.

#If it doesn't work

• "Edit Config" isn't there — update Claude Desktop to a current version.
• Tools don't appear after saving — fully quit and relaunch (not just close the window), and check the JSON is valid (a stray comma breaks it).
• npx errors / "command not found" — install Node.js 18+ and reopen Claude Desktop so it picks up your PATH.
• Calls return an auth error — that endpoint needs full content; add SKILLDB_API_KEY in the env block (Step 2) with a free key from skilldb.dev/api-access.
• The model ignores SkillDB — nudge it: "use the SkillDB tools to…". After a use or two it reaches for them on its own.

#Other agents work too

The same package works with any MCP client:

• Claude Code — claude mcp add skilldb -- npx -p skilldb skilldb-mcp
• Cursor — add the same JSON config under Settings → MCP.
• OpenClaw — npx skilldb init detects OpenClaw and wires it up.
• Hermes — hermes config mcp.add skilldb -- npx -p skilldb skilldb-mcp.

The /mcp page has per-agent tabs with exact steps.

#What you get for free

The free tier covers most individual use — search, browse, and read full skill content up to 50 calls/day, no card. Hitting the limit (using it on every PR, or in CI)? Pro is $9/mo for 10,000 calls/month; Studio is $29/mo for 50,000 calls plus authoring your own private skills via skilldb_create_skill.

#What's inside

The catalog right now: 5,900+ skills across 428 packs and 37 categories (live count at skilldb.dev/api/stats). A short tour:

• Software engineering — code review, security practices, system design, performance, debugging
• Incident response — blameless postmortems, severity classification, runbooks, customer comms
• DevSecOps — SAST/DAST, supply-chain security, threat modeling, security as code
• AI / agent engineering — prompt-injection defense, multi-agent orchestration, agent memory
• Domain expertise — grant writing, healthcare informatics, contract review, financial analysis

Each skill is 200–500 lines of focused markdown — short enough to fit in context, long enough to change the agent's output.

#The big picture

MCP lets any agent be extended by anyone, without forking the model or rewriting the client. SkillDB's bet is that the highest-value thing to extend an agent with is expert knowledge — the accumulated practice of specialists across thousands of domains. Install it once; every conversation gets better.

→ https://skilldb.dev/mcp