Skip to main content

Three surfaces for your team's skill lifecycle

MCP for any agent. CLI for the terminal. VS Code for the editor.

One local skill database. One API key. Three surfaces over the same registry — author, distribute, govern, and retire skills from wherever your team already works. (Throughout this page, MCPmeans the Model Context Protocol — the standard agents like Claude Code, Cursor, Copilot, Codex, and Windsurf use to talk to external tools.)

Capability comparison

Capabilities marked via MCP for VS Code mean the extension actively spawns the MCP server to deliver them — not every tool the MCP server happens to expose. Team workspaces, audit export, SIEM, and RBAC are MCP-only and have no VS Code UI affordance today.

CapabilityMCPCLIVS Code
Search / get / install / uninstallyesyesvia MCP
In-session contextual recommendation (skill_suggest)yesNot supportedvia MCP
Side-by-side compare with LLM-readable rationaleyesNot supportedvia MCP
Inline install during an agent chat sessionyesNot supportedNot supported
Sidebar tree, quick-pick UI, rendered detail panelNot supportedNot supportedyes
SKILL.md frontmatter intellisense + diagnosticsNot supportedNot supportedyes
Create-skill wizard (4-step, GUI)Not supportedpartialyes
Skill authoring (init / validate / subagent / transform / mcp-init)Not supportedyesNot supported
Publish to registry (author publish)Not supportedyesNot supported
CI-safe exit codes + JSON outputNot supportedyesNot supported
Offline local search after syncNot supportedyesNot supported
Device-code login (RFC 8628)Not supportedyesNot supported
Team workspaces / audit export / SIEM / RBACyesNot supportedNot supported
Single API key works across all threeyesyesyes

Working in an MCP-compatible agent (Claude Code, Cursor, Copilot, Codex, Windsurf)? Use the MCP server.Scripting, CI, or authoring? Use the CLI.Living in VS Code? Use the extension.

MCP server

@skillsmith/mcp-server

Pick this if: Working in any MCP-compatible agent (Claude Code, Cursor, Copilot, Codex, Windsurf).

Read the docs →

CLI

@skillsmith/cli

Pick this if: Scripting, CI, or authoring your own skills.

Read the docs →

All three surfaces read the same local SQLite database at~/.skillsmith/skills.db. Search runs against an FTS5 full-text index (SQLite's built-in keyword search) by default; semantic search is opt-in and uses local ONNX embeddings (an open ML model format that runs on CPU, no API call) — your queries never leave your machine.

Read the deep dive →

More questions about which surface to use, what gets cached locally, or howsync works? See theTechnical FAQ.