Submit a skill

Paste a public GitHub repository URL or a link to a SKILL.md file. Leave the revision empty to use the default branch. Scan options let you specify a branch, tag, commit SHA, or skill path.

For branch names containing slashes, use the repository root URL and enter the complete branch name in Scan options. The scanner resolves an exact commit before inspecting the bundled files.

Scan MCP server source

Choose MCP server on the homepage and paste a public GitHub repository or folder URL. Scan options let you select a server folder and revision. No SKILL.md is required. Static checks inspect tool, resource, and prompt declarations, transports, and capabilities alongside NOVA/YARA. Capability signals are separate from correlated findings. The server is never started or connected.

Choose Upload MCP to inspect a ZIP, .mcpb, or legacy .dxt source bundle up to 2 MiB, with the same 200-file, 1 MiB-per-file, and 10 MiB expansion limits as skill ZIPs. The scanner extracts the bundle and inspects its contained files with NOVA/YARA and the configured static checks. No SKILL.md is required. AI assessment is optional for GitHub and uploaded MCP source; required secret checks cover the source inventory. Unsupported files, declaration limits, or checks that cannot complete remain coverage gaps. No MCP server is installed, started, or connected, and dependencies and runtime behavior are not verified. Submitted reports stay unlisted and never populate public discovery.

Upload a skill

Choose Upload skill on the homepage. Upload a UTF-8 SKILL.md file (up to 1 MiB), or a ZIP bundle ending in .zip or .skill (up to 2 MiB). Each skill folder must contain SKILL.md, with its related files underneath. ZIPs allow up to 200 files, 1 MiB per file, 10 MiB unpacked, and 20 skills. Links, encrypted transport entries, and unsafe paths are rejected. Nested archives are passed to the scanner for bounded inspection; unsupported or uninspected contents remain coverage gaps.

Uploaded files are inspected without execution and removed after processing. Upload reports show a SHA-256 hash; they do not establish a GitHub origin. Uploads use the same verification and shared scan allowance as URL submissions.

Read and share the result

Keep the report link while the scan runs. The finished report includes a source summary, coverage, correlated findings, raw rule matches, and requested AI assessment. MCP reports also show statically observed capabilities and declaration coverage. A capability describes what the inspected source can do; its severity is separate from a maliciousness verdict. Share the unlisted link or export the JSON. An incomplete scan has its own outcome even when some checks succeeded.

Add AI context

Select Add AI intent assessment when submitting a scan. AI is optional and advisory; it does not replace the rule checks. Requested AI that cannot complete makes the scan incomplete.

Browse skills and MCPs

The public scan archive contains separate observer results for agent skills and MCP server source. Choose All types, Agent skills, or MCP servers, then filter by discovery source, findings, AI assessment, date, or score. The MCP discovery source includes GitHub sources and registry listings; it does not mean a registry package or running endpoint was tested.

Click column headings to sort all matching observations, then browse pages of 25 or 10 results. The archive refreshes every 30 seconds while visible, preserving your filters and page. Use “Newest observations” to return to the latest scans. Discovery runs every 15 minutes in bounded batches; it is not a complete index of all skills, MCP servers, or bundles. Results publish at batch completion. Submitted reports and uploads do not enter this archive.

Integrate scans

Use the keyless REST API or add the LOLSkills scan skill to Claude Code or Codex. Both share the public service allowance. No infrastructure setup or API key is needed.

Join the community

Visit Community & code access to connect with the MagicSword team, discuss integrations, or request access to the private repository.

Understand the scope

Read the detection methodology and Privacy & sharing before submitting a scan.