Career Navigator

An MCP (Model Context Protocol) server that exposes the career diagnosis AI system as callable tools — powered by Google Gemini and a RAG knowledge base of 10 professions.

Any MCP-compatible client (Claude Desktop, Cursor, etc.) can integrate this server and call its tools directly.

---

Tools

analyze_skills

Analyzes free text describing a user and returns a list of skills with match percentages.

Input:

{ "text": "I enjoy working with people and solving complex problems..." }

Output:

[
  { "skill": "Interpersonal Communication", "match_percentage": 95 },
  { "skill": "Problem Solving",             "match_percentage": 85 }
]

---

recommend_profession

Receives a skills array and recommends the most suitable profession using RAG — matched against a real knowledge base of 10 professions.

Input:

{
  "skills": [
    { "skill": "Interpersonal Communication", "match_percentage": 95 },
    { "skill": "Problem Solving",             "match_percentage": 85 }
  ]
}

Output:

[
  {
    "profession": "מנהל משאבי אנוש",
    "match_percentage": 91,
    "explanation": "..."
  }
]

---

Architecture

MCP Client (Claude Desktop / Cursor / any AI agent)
    │
    │  tools/call: analyze_skills / recommend_profession
    ▼
career-navigator-mcp (StdioServerTransport)
    │
    ├── analyze_skills ──► Gemini API (structured JSON output)
    │
    └── recommend_profession ──► RAG: professions_data.json
                                      └── top 3 matches injected into Gemini prompt

---

Setup

Prerequisites

• Node.js 18+
• A GEMINI_API_KEY from Google AI Studio

Install

git clone https://github.com/Miriam-Epstein/career-navigator-mcp.git
cd career-navigator-mcp
npm install

Environment

Create a .env file in the project root:

GEMINI_API_KEY=your_key_here

---

Connecting to Claude Desktop

Add this to your claude_desktop_config.json:

{
  "mcpServers": {
    "career-navigator": {
      "command": "node",
      "args": ["/absolute/path/to/career-navigator-mcp/index.js"]
    }
  }
}

---

Model Fallback

Every Gemini call uses an automatic fallback chain:

gemini-2.5-flash-lite  →  (on 429 / quota error)  →  gemini-2.5-flash

---

Tech Stack

Layer	Technology
Protocol	MCP SDK (@modelcontextprotocol/sdk)
Transport	StdioServerTransport
AI	Google Gemini (@google/genai)
RAG	Local JSON knowledge base (10 professions)