SkillMD MCP server
Featured official server
5 tools · Last check that passed 4 Nov 2026, 12:15 UTC
SkillMD's own remote MCP server gives an agent search across SkillMD's registry of Agent Skills, SKILL.md files written by their own authors and mostly indexed from public repositories, with each skill's provenance, capability flags and full files to download, with no account. Once a day, Stackduke connects to it the way any MCP client would, with no account and no sign-in, and lists its tools. It never runs them.
All 5 tools
| Tool | Counted as |
|---|---|
skillmd_download | Said to only read |
skillmd_get | Said to only read |
skillmd_recommend | Said to only read |
skillmd_search | Said to only read |
skillmd_trending | Said to only read |
The tools the server listed at the last check that passed, at its default address and settings, with no sign-in; an account that signs in, or other settings, may see other tools. What a tool is counted as doing is Stackduke's cautious reading of the server's labels, the tool's name and its description: where they disagree it takes whichever says the tool can do most, so it overcounts on purpose. It does not say the server is safe, and it does not list what a tool cannot do. Descriptions are not shown, because they are written for models and may try to steer them.
Recent checks
21 of the last 21 checks passed
A check passes when the server lists its tools. Checked at https://api.skillmd.com/mcp.
Recent changes
No change seen across the last 21 checks that passed.
From the scan of 4 Nov 2026, 12:43 UTC. "Changed" and "Can do more" count changes from the last 30 days; older ones stay listed here. A change made and undone between two checks is not seen. Every check is in the scanner's public data.
Know when SkillMD's tools change.
One email when SkillMD's official MCP server adds, removes or changes tools, with what changed. Nothing else.
What to watch for
- SkillMD says a skill is instructions an agent follows with your permissions, sometimes with scripts beside it, and that nothing sandboxes the words. skillmd_download returns a skill and its companion files, ready to write to disk. SkillMD's page on skill security
- SkillMD says its capability scan does not decide whether a skill is safe, and that on 17 September 2026 only 51.9% of its public skills carried capability flags and 4.5% a third-party scanner result. SkillMD's page on skill security
- SkillMD's terms call its safety review best-effort, with no guarantee that a skill is free of defects, vulnerabilities or unwanted behavior. SkillMD's terms of service
- Skill files written by third parties come back to the agent as text, and text an agent reads can carry instructions for it. OWASP, prompt injection
How to limit it
With SkillMD's own settings, as its documentation describes them. Stackduke has not tested them.
- We recommend searching with verified_only, which SkillMD says returns only skills that passed its safety review. Its terms say that review is best-effort. SkillMD's MCP documentation
- Read a skill, and run SkillMD's open-source scanner on it, before installing it, as SkillMD advises. SkillMD's page on skill security
- SkillMD says the remote server needs no account, so we recommend connecting it with no SkillMD credentials. SkillMD's MCP server page
What Stackduke is designed to do
Stackduke is designed to sit between your agents and their MCP servers: to check each tool call against your rules before it runs, hold the risky ones for a person to approve, and keep a record of what was decided.
It does not yet forward or block tool calls for customers. Today it reads what this server offers and publishes what changes.