Portfolio
Model Context Protocol · agent-native CV

Most candidates put their CV on a website.
I put mine in a protocol.

If you're hiring AI engineers in 2026, you probably use AI assistants. Why not let your assistant interview my CV directly — structured, grounded, citation-friendly — instead of copy-pasting my résumé into a prompt?

This is a working Model Context Protocol server. Six tools, two resources. Point your MCP-compatible agent (Claude Desktop, Cursor, Cline, etc.) at it and ask questions.

Tools exposed

  • get_profile

    Name, title, location, availability, contact, book-a-call link.

  • list_projects

    Flagship projects with slugs and one-line impact metrics.

  • get_project(slug)

    Full structured case study for one project.

  • list_skills

    Categorised skills — Gen AI, RAG, frameworks, cloud, full-stack.

  • search_cv(query)

    Keyword search across the corpus — returns ranked chunks.

  • check_fit

    Hands off JD analysis to the live /fit analyser.

Plus 2 resources · cv://summary (plain text) · cv://corpus (full JSON)

Wire it up in two minutes.

  1. 1

    Clone the repo

    git clone https://github.com/mazhar1790/portfolio
    cd portfolio/scripts/mcp-server
    npm install
  2. 2

    Smoke-test the server

    npm test
    # Should print: ◆ Response #1, #2 … listing tools and a profile JSON.
  3. 3

    Add to Claude Desktop config

    Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

    {
      "mcpServers": {
        "mazhar-cv": {
          "command": "node",
          "args": [
            "/absolute/path/to/Portfolio/scripts/mcp-server/server.mjs"
          ]
        }
      }
    }

    Restart Claude. The mazhar-cv tools appear in the tools palette.

  4. 4

    Ask your agent

    "Use the mazhar-cv tools to tell me about his RAG production experience and whether he's available."

Read-only

No writes anywhere. Safe to point any agent at it.

Zero hallucination

Tools return structured data. Agents quote chunks, not invent them.

Reusable pattern

MIT-licensed. Fork it. Adapt it. Make your own agent-native CV.

This is the kind of move I bring to your team.

Recognising MCP early. Shipping a working integration in an afternoon. Documenting it well enough that anyone can fork it. That's the bar.