MCP Tool Reference

Detailed specification for the available AI Agent tools and expected JSON schemas.

Available Tools

Our MCP server currently exposes one primary tool for autonomous AI agents.

---

find_candidates_by_skill

This tool allows an AI agent to scan a local directory of PDF resumes, pre-filter them for a specific technical keyword using Python, and extract deep structured data using Gemini 2.5 Flash.

Input Parameters

Agents must provide the following arguments when calling this tool:

skill · string · required

The specific technical skill to search for (e.g., “Java”, “Python”, “Spring Boot”).

directory_path · string · default: "./resumes"

The local folder path where the PDF resumes are stored.

Expected Output

The tool enforces strict data validation via Pydantic. It returns a structured JSON array of candidates matching the requested skill.

candidates · array

Hide Candidate Object Properties

full_name · string

The candidate’s extracted full name.

role · string

The candidate’s current or targeted job title.

phone_number · string

The extracted contact number.

top_5_technical_skills · array of strings

A list of the candidate’s top technical proficiencies.

---

Testing with an Agent

If you are testing this integration inside an MCP-compatible client (like Claude Desktop), you can trigger the pipeline using a natural language prompt like this:

“Please use the find_candidates_by_skill tool to find all candidates who know ‘Java’ in the resumes folder. Present the results to me in a clean markdown table.”