skill-router
OfficialClick on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@skill-routersearch for a web scraping skill"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
skill-router
A lazy router for Claude Code skills: instead of every installed skill costing context every turn, expose the whole library through three small MCP tools — search, load, reindex. Weak/cheap models navigate it fine; that's the point.
Single-file Node MCP server (stdio). One dependency (@modelcontextprotocol/sdk).
No database, no embeddings, no background services. Indexing ~185 skills takes
~25ms at startup.
How it works
On startup (and on skill_reindex) the server walks three roots for SKILL.md
files:
Root | Label | Notes |
|
| symlinks followed (with cycle guard) |
|
|
|
|
| see below |
Frontmatter name:/description: are parsed (including YAML block scalars
like description: >-); duplicates dedupe with precedence
active > plugin > vault.
Tools
skill_search{query, limit?}— ranked{id, name, description, source, path, score}. Token-overlap scoring with query-side stopword filtering (so "make my website load faster" still findsweb-perf), whole-query substring boost, and partial name-fragment credit ("website" matches the nameweb-perf).skill_load{id}— fullSKILL.mdcontent plus sibling files (references/,scripts/, …). Accepts ids or names case-insensitively; unknown ids return the 5 closest matches instead of an error.skill_reindex— rescan, returns counts per source.
The vault
The real context win: move rarely-used skills from ~/.claude/skills/ into
~/.claude/skill-vault/. The harness stops listing them every turn, but they
stay fully discoverable and loadable through the router.
Related MCP server: claude-skills-mcp
Install
npm install
claude mcp add --scope user skill-router node /absolute/path/to/server.mjsTest
node test.mjsSelf-contained: builds a fixture skill tree in a temp HOME (including a
symlinked skill, a block-scalar description, an excluded plugin-cache entry,
and a shadowed vault duplicate) and asserts against it over real stdio JSON-RPC.
Security note
The indexer follows symlinks out of the skill roots and will index any
SKILL.md they reach — so don't point skill roots at (or install skill
bundles containing symlinks to) untrusted or sensitive locations. Loaded
skill content is returned as data, never executed.
Credits
The scoring approach is ported from the Hermes lazy-router plugin in agency-agents (MIT, AgentLand Contributors) — the pattern of exposing a large roster as a few search tools instead of preloading everything.
License
MIT
This server cannot be installed
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Latest Blog Posts
- Your AI Chatbot Just Exposed Your CEO's Salary to an InternBy Om-Shree-0709 on .Agent IdentityMCP SecurityOAuth Delegation
- Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)By Om-Shree-0709 on .Agentic AiPrompt InjectionWebAssembly
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/sorcerai/skill-router'
If you have feedback or need assistance with the MCP directory API, please join our Discord server