Resume Assistant

Architecture

System Flow

graph TD
    A[User / Agent] -->|Calls Tool| B(MCP Server)
    B -->|parse_resume_local| C{Local Parser}
    C -->|Reads PDF/DOCX| D[(Local Disk)]
    B -->|fetch_job_description| E{Jina Scraper}
    E -->|Fetches URL| F((Job Board/LinkedIn))
    B -->|Analytical Tools| G[Heuristic Engine]
    C --> G
    E --> G
    G -->|Returns JSON| A

Sequence: Resume → Job URL → ATS Analysis

sequenceDiagram
    participant U as User / Agent
    participant M as MCP Server
    participant FS as File System
    participant Web as Job Board
    
    U->>M: parse_resume_local(file_path)
    M->>FS: Read PDF/DOCX
    FS-->>M: Raw Resume Text
    M-->>U: Return Text
    
    U->>M: fetch_job_description(url)
    M->>Web: Scrape Job Text
    Web-->>M: Raw Job Text
    M-->>U: Return Text
    
    U->>M: calculate_ats_match(resume_text, job_text)
    M->>M: Compute Keyword Overlap & Score
    M-->>U: ATS Match Results & Missing Skills

Features

Tool	Description
parse_resume_local	Extracts raw text securely from a local PDF, DOCX, TXT, or MD file.
fetch_job_description	Scrapes the raw text of a job posting from a URL (e.g. LinkedIn, Greenhouse).
calculate_ats_match	Calculates keyword overlap and generates a percentage ATS match score.
generate_interview_questions	Generates targeted technical and HR interview questions based on match results.
get_resume_improvements	Analyzes resume content (length, metrics, action verbs) and provides actionable tips.
generate_cover_letter	Generates a targeted cover letter template injecting candidate-specific matches.
linkedin_profile_optimizer	Analyzes LinkedIn profile text and suggests formatting and content improvements.
skill_gap_analysis	Compares resume against job to identify missing skills and generate a learning roadmap.

Quick Start

1. Clone the repository:

   git clone https://github.com/rajesh-kayal-dev/resume-assistant-mcp.git
   cd resume-assistant-mcp
   

2. Install dependencies:

   npm install
   

3. Start the development server:

   npm run dev
   

4. Run tests:

   npm test
   

5. Test with MCP Inspector (or MCPize Playground):

   npx @modelcontextprotocol/inspector node dist/index.js
   # Or using MCPize:
   mcpize dev --playground
   

Example Prompts

Copy and paste these prompts into your favorite MCP-compatible AI (like Claude Desktop or the MCPize Playground):

1. "Parse my resume at C:\resumes\my_resume.pdf."
2. "Fetch the job description from https://www.linkedin.com/jobs/view/12345/."
3. "Parse my resume at resume.pdf and compare it against https://example.com/job. Give me the ATS match score."
4. "Analyze my resume at resume.pdf and provide improvement tips."
5. "Read my resume at resume.pdf and the job at https://example.com/job, then generate a custom cover letter."
6. "Run a skill gap analysis for my resume at resume.pdf against the job description at https://example.com/job."
7. "Generate 5 technical interview questions based on my missing skills for the job at https://example.com/job using my resume resume.pdf."
8. "Optimize my LinkedIn About section text: 'I am a software engineer looking for a job...'"

API Reference

parse_resume_local

Extract raw text from local files. Input:

Parameter	Type	Required	Description
file_path	string	Yes	Absolute or relative path to the local resume file.

Output:

{
  "text": "John Doe\nSoftware Engineer..."
}

Note: Uses unpdf and mammoth for local parsing without external API calls.

fetch_job_description

Scrape job postings from URLs. Input:

Parameter	Type	Required	Description
url	string	Yes	The full URL of the job posting.

calculate_ats_match

Compute ATS Match Score. Input:

Parameter	Type	Required	Description
resume_text	string	Yes	The raw parsed text of the candidate's resume.
job_text	string	Yes	The raw text of the job description.

Output:

{
  "match_score": 85,
  "matched_keywords": ["react", "typescript"],
  "missing_keywords": ["docker", "kubernetes"]
}

skill_gap_analysis

Generate a multi-phase learning roadmap. Input:

Parameter	Type	Required	Description
resume_text	string	Yes	Raw resume text.
job_text	string	Yes	Raw job text.

Project Structure

resume-assistant-mcp/
├── src/
│   ├── index.ts                 # MCP Server entry point and tool registration
│   ├── tools/                   # Pure function logic for all 8 tools
│   │   ├── parse-resume-local.ts
│   │   ├── calculate-ats-match.ts
│   │   └── ...
│   └── lib/                     # Shared utilities
│       ├── api-client.ts        # Cached scraper
│       └── cache.ts             # In-memory caching logic
├── tests/                       # Vitest unit tests
├── package.json                 # Dependencies and scripts
├── mcpize.yaml                  # Deployment configuration
└── test-mcp.sh                  # Smoke test script

Deployment

Deploy via MCPize

mcpize deploy

Your server will be instantly available on the MCPize Cloud.

Bare Metal / Docker

You can easily containerize this application by building a standard Node.js Docker image, exposing port 8080, and running npm start.

Roadmap

• Short-term: Improve stopword filtering in the ATS matching heuristic.
• Medium-term: Add native PDF generation to export customized cover letters.
• Long-term: Implement local embeddings/RAG for deeper semantic skill matching instead of exact keyword matching.

Use Cases

• Job Seekers: Automatically tailor resumes and cover letters for hundreds of applications.
• Career Changers: Identify exact skill gaps to bridge between current experience and target roles.
• Students: Get actionable formatting feedback on their first professional resumes.
• Recruiters: Batch-analyze candidates against specific job descriptions to rank them by objective ATS criteria.
• Career Coaches: Generate automated reports and interview prep questions for clients.

FAQ

Q: Does this cost money to run? A: No! The server uses a Zero API Cost strategy. All ATS scoring, parsing, and analysis rely on local regex and string manipulation heuristics rather than expensive LLM endpoints.

Q: Is my resume data sent to OpenAI? A: No. parse_resume_local reads the file locally on your machine. The text is only sent to an LLM if the AI client you connect the server to decides to pass it in a prompt.

Q: Why isn't my PDF parsing correctly? A: Ensure it is a text-based PDF. Image-based PDFs (scanned documents) are not currently supported by unpdf.

Q: Why does the job scraper fail on some sites? A: The scraper uses the Jina Reader API, which handles JavaScript-heavy sites well, but certain extremely strict bot-protected sites may block it.

Q: Can I run this without MCPize? A: Absolutely! It is a standard Express-based HTTP server exposing the official @modelcontextprotocol/sdk.

Q: How accurate is the ATS Match score? A: The heuristic engine strips stopwords and matches alphanumeric keywords. It is a highly effective approximation of older ATS systems but lacks deep semantic understanding (e.g., it won't map "Node.js" to "JavaScript" automatically). This limitation is documented and will be addressed in future roadmap updates.

Q: How do I change the port? A: Set the PORT environment variable before running the server (e.g., PORT=3000 npm run dev).

Q: Can I add my own tools? A: Yes! Simply create a new file in src/tools/, write a pure TypeScript function, define the Zod schema, and register it in src/index.ts.

---

Built with MCPize