Skip to main content

​
Overview

Cursor discovers skills directly from the skills directory (.claude/skills/ by default) and via the FastSkill MCP server. FastSkill’s job is to install the right skills onto disk and keep them reconciled with your manifest — there is no metadata file to generate or keep in sync.
Modern agents (Cursor, Claude Code, …) read installed skills from the skills directory natively. Just fastskill install (or fastskill add) and the skills are available. Register the MCP server below for richer, tool-based access from inside Cursor.

​
Quick start

# 1. Install the skills declared in skill-project.toml
fastskill install

# 2. Verify what is on disk
fastskill list

# 3. (Optional) Build the local index for semantic search
fastskill reindex --skills-dir .claude/skills/

​
MCP registration

Register fastskill as an MCP server inside Cursor with a single command: 
# Register for the current project (stdio transport, recommended)
fastskill mcp install --agent cursor --stdio --scope project --overwrite

# Preview the config change without writing any files
fastskill mcp install --agent cursor --stdio --dry-run
After running the command, reload Cursor. The tools exposed by fastskill mcp serve --transport stdio will be callable from the agent.
Deprecated: Manually editing .cursor/mcp.json to add a fastskill MCP entry is deprecated. Use fastskill mcp install instead.

​
How Cursor uses the skills

When you ask Cursor to perform a task, it:
  1. Reads available skills from the skills directory (and the FastSkill MCP server, if registered).
  2. Matches task requirements against each skill’s description frontmatter.
  3. Loads the full SKILL.md for relevant skills to get instructions and parameters — the same content fastskill read <skill-id> prints.

​
Example

User query: “Help me create a PowerPoint presentation for my quarterly review.” Cursor scans installed skills for presentation-related descriptions, finds a match (e.g. pptx), loads its SKILL.md, and follows the skill’s instructions to complete the task.

​
Configuration

Ensure your project has a skill-project.toml and that Cursor and FastSkill agree on the skills directory. The default is .claude/skills/; override it via skills_directory in .fastskill/config.yaml or the --skills-dir flag. 
fastskill install
fastskill list

​
Multiple skill directories

fastskill reindex --skills-dir .claude/skills/
fastskill reindex --skills-dir ./custom-skills/

​
Automation

Keep the skills directory reconciled in CI so commits always carry the resolved skill set: 
# .github/workflows/reconcile-skills.yml
name: Reconcile skills
on:
  push:
    paths:
      - 'skill-project.toml'
      - 'skills.lock'

jobs:
  install-skills:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - name: Install FastSkill CLI
        run: |
          # Add the steps your org uses to place `fastskill` on PATH (release tarball, package, etc.)
          fastskill -V
      - name: Install skills from lock
        run: fastskill install --lock
      - name: Verify
        run: fastskill list

​
Troubleshooting

Cursor doesn’t see a skill 
# Confirm it is installed and reconciled
fastskill list

# Confirm its content is readable
fastskill read <skill-id>
Semantic search returns nothing 
# Reindex (requires an embedding provider; see `fastskill doctor`)
fastskill reindex --skills-dir .claude/skills/

# Keyword fallback works without a provider
fastskill search "test query" --local --embedding false
Check environment readiness 
fastskill doctor

​
Summary

FastSkill’s Cursor integration provides:
  • Direct skill discovery — Cursor reads installed skills from the skills directory, no sync step
  • MCP access — register the FastSkill MCP server for tool-based access inside Cursor
  • Reproducible installs — skill-project.toml + skills.lock keep the skill set consistent across the team
  • Optional semantic search — fastskill reindex + fastskill search --local when an embedding provider is configured