Skip to content
MCP server

Skill Shelf (Rust singleton daemon)

Local skill library served as 7 MCP tools so agents load skills on demand

Platforms
CLI, macOS, Windows
Version
2.0.5
Listing updated
today

Similar MCP servers

About Skill Shelf (Rust singleton daemon)

Skill Shelf (Rust singleton daemon) is an MCP server for agent tooling. It is available on CLI, macOS, Windows and Linux.

The README says Skill Shelf keeps a large collection of skill files in a local repository and exposes only seven MCP tool definitions to the client, so the model searches and loads a skill only when a task calls for it. Each skill is a Markdown file with frontmatter, and skills are grouped into named shelves; the bundled set includes 524 skills across 18 groups.

A Rust daemon holds the index and does the work, while a small Node stdio shim is started by each MCP client and forwards requests to it. The README states that Claude Code, Claude Desktop, Cursor and Windsurf can share one daemon process, and that the index hot-reloads when files in the packages folder change.

The project suits people who want to keep many specialised prompts or team know-how available to an agent without paying their token cost on every conversation. It also lets users install their own skill packs and create custom groups with the manage_group and install_skills tools.

Tools this server provides (7)

Taken from the project documentation; the running server may differ.

browse_shelf
Returns the group catalog, or the skill summaries for a given group.
search_skills
Keyword search across all skills, returning the top matches.
read_skill
Reads a skill summary, or the full body, references and resources with full=true.
install_skills
Installs a skill package from a directory, with optional LLM-assisted grouping.
validate_skills
Checks skill integrity and can remove problematic skills when clean=true.
manage_group
Creates, updates or deletes storage groups.
reclassify_skill
Moves a skill to a different group and rebuilds the index.

What it does

Progressive group routing
Browse a group catalog first, then the skills inside a chosen group, and fall back to keyword search only when needed.
Seven-tool surface
Three read tools and four write tools keep the tool definitions in the client context small.
Shared Rust daemon
One background process serves all connected MCP clients, avoiding a separate process per client.
Custom skill packs
Install a directory of Markdown skills and validate it, with an option to remove invalid entries.
Hot reload
File watching on the packages folder updates the index when skills are added, changed or deleted.
Bilingual search
Searches try the user's language first and retry in English when there are no results.

Good for

  • Give an agent access to hundreds of specialised skills without loading them all into context
  • Turn a team's recurring workflows and lessons learned into a shared, searchable skill library
  • Share one skill index across several MCP clients on the same machine
  • Organise custom skill groups and reclassify skills as a library grows

Skill Shelf (Rust singleton daemon) FAQ

How many tools does the server expose?

The README lists seven tools: browse_shelf, search_skills and read_skill for reading, and install_skills, validate_skills, manage_group and reclassify_skill for writing.

Does it need Rust installed?

No. The README says release packages include a prebuilt Rust daemon binary for each supported platform, so Rust and cargo are not needed for that install route.

Where are the skills stored?

In a local data/hub directory, with packages organised as group folders containing SKILL.md files. The SKILL_SHELF_ROOT variable can point the library elsewhere.

Is it free to use?

The README does not state a licence or pricing for the project itself. It credits several MIT-licensed open source projects as sources of some bundled skill content.

What are the best alternatives to Skill Shelf (Rust singleton daemon)?

The closest alternatives to Skill Shelf (Rust singleton daemon) are sciver and Meta Skill.

Where this comes from

  • Description: written from the project's README and its MCP Registry entry.

How AIPoweredBot checks servers