Four-Leaf MCP server
The open job-search and interview-prep MCP. Works in any MCP-aware AI assistant. Hand off to Four-Leaf for live voice mock interviews and AI resume tailoring.
Free Four-Leaf account required (the 3-day trial works). Browser-based OAuth login on first use. No API keys.
What you get
list_roles— the catalog of roles Four-Leaf has interview intelligence forget_role_intelligence— typical interview pipeline, experience-level calibration, 5-dimension scoring rubric, resume + cover-letter guidance for any roleget_interview_questions— pull questions from the Four-Leaf bank by role, difficulty, and typegenerate_practice_questions— fresh questions on demand for a role, type, and difficultysearch_jobs— natural-language search over 180k+ active job postingsmatch_score— score a resume against a job description with skill-gap analysisexplain_interview_format— grounded walk-through of what to expect, how to win, and red flagscomp_coach— analyze a job offer and get a negotiation strategy with talking pointscomp_benchmarks— a market salary band for any role, grounded in a live web search with sourcesstart_voice_mock_interview(paid) — deep-link into a voice mock with adaptive follow-ups and rubric-scored feedbacktailor_resume(paid) — deep-link into AI resume tailoring with your role and JD context surfaced
Two install paths
Most users want one or both of these. Pick whichever fits your editor.
1. Hosted MCP (works in any MCP-aware client)
Connect the hosted endpoint directly. Works in Claude Desktop, Claude Code, Cowork, Cursor, ChatGPT Desktop, Perplexity, Cline, Continue, and Windsurf. Browser-based OAuth on first call. No API keys.
One-line install for Claude Code:
claude mcp add four-leaf --transport http https://four-leaf.ai/api/mcpPer-client snippets for everything else are below.
2. Open-source Skill (works in 17+ editors)
The MIT-licensed Skill wrapper adds a coaching layer on top of the MCP tools. Routes seven workflows (kickoff, find-jobs, prep-role, practice, analyze-jd, negotiate-prep, interview-strategy) based on what you ask for. Source at github.com/fourleafai/clover-public.
One command for any skills-aware editor (Claude Code, Cursor, Codex, Cline, Amp, OpenCode, Zed, Gemini CLI, and 9 more):
npx skills add fourleafai/clover-publicOr, for tool-specific bundles (Claude Code, Cursor, OpenAI Codex CLI, GitHub Copilot):
npx four-leaf-coach addThe Skill is useful but not required. The MCP tools work on their own.
Per-client install (hosted MCP)
The hosted MCP lives at https://four-leaf.ai/api/mcp. Drop the snippet for your client into its config, restart, and complete the OAuth login the first time your assistant invokes a tool.
Claude Desktop
Config path: ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%/Claude/claude_desktop_config.json (Windows)
{
"mcpServers": {
"four-leaf": {
"url": "https://four-leaf.ai/api/mcp"
}
}
}Restart Claude Desktop after editing. On first tool call, your browser will open for Four-Leaf login.
Claude Code (terminal)
Config path: Run from any project directory
claude mcp add four-leaf --transport http https://four-leaf.ai/api/mcpOne command. OAuth opens in your default browser the first time the MCP is invoked.
Claude Cowork
Config path: Customize → Connectors → + (add custom connector)
Name: Four-Leaf
URL: https://four-leaf.ai/api/mcpCowork handles the OAuth flow in-app. Once connected the 11 tools show up under Tool permissions.
Cursor
Config path: ~/.cursor/mcp.json
{
"mcpServers": {
"four-leaf": {
"url": "https://four-leaf.ai/api/mcp"
}
}
}Reload Cursor after editing. OAuth flow opens in browser.
ChatGPT Desktop
Config path: Settings → Apps & Connectors → Advanced → Developer Mode → Add
Name: four-leaf
URL: https://four-leaf.ai/api/mcpRequires ChatGPT Plus/Pro/Business/Enterprise/Education with Developer Mode enabled. Free ChatGPT users cannot install MCPs at this time.
Perplexity
Config path: Settings → Connectors → + Custom
Name: Four-Leaf
URL: https://four-leaf.ai/api/mcpRequires Perplexity Pro, Max, or Enterprise. OAuth opens in browser on first call.
Cline (VS Code)
Config path: Cline settings → MCP servers → Add Server
Server URL: https://four-leaf.ai/api/mcpCline handles the OAuth dance in the VS Code window.
Continue (VS Code / JetBrains)
Config path: ~/.continue/config.json
{
"experimental": {
"modelContextProtocolServers": [
{
"transport": { "type": "streamable-http", "url": "https://four-leaf.ai/api/mcp" }
}
]
}
}Windsurf
Config path: Settings → MCP servers → Add
URL: https://four-leaf.ai/api/mcp
Type: streamable-httpManage connected apps
See all the AI assistants you've connected, and revoke any of them, at Settings → Connected apps.
Self-host (advanced)
The reference implementation lives at github.com/fourleafai/clover-public. MIT-licensed. Useful if you want to read the code, fork it, or run a local stdio version without auth. Most users want the hosted version above.
FAQ
Skill or MCP? What's the difference?
The MCP server is the tools (search_jobs, match_score, comp_benchmarks, etc.). It runs on our servers and any MCP-aware client can call into it. The Skill is a thin routing layer that runs in your editor and tells the agent when to call which tool. You can install just the MCP and your AI will figure out the rest. Most users do.
Does this work with free ChatGPT?
No. OpenAI gates MCP-aware tools behind Developer Mode on paid ChatGPT plans. Free ChatGPT users can't install MCPs (yet).
Do I need a paid Four-Leaf account?
Most tools work on a free Four-Leaf account (the 3-day trial works). Two tools require an active paid plan: the voice mock interview kickoff and AI resume tailoring.
What data does Four-Leaf see when I use the MCP?
Only the arguments your AI assistant passes to our tools, plus your account identity. We do not see the rest of your chat. Standard Four-Leaf privacy policy applies.
How do I revoke access?
Settings → Connected apps. Revoking takes effect immediately on the next tool call.