Skip to main content
Glama

SkillMatch MCP

Claude-powered job fit analyzer for people who prove their skills through work, not credentials.

What It Does

SkillMatch is an MCP server that gives Claude access to your GitHub portfolio, resume, and job search preferences. It lets Claude:

  • Analyze how well you fit a specific job description based on your actual projects and resume

  • Build optimized job search queries tailored to your targets and constraints

  • Track every application you submit in a local SQLite database

  • Pull your public GitHub repos to highlight relevant work during fit analysis

All data stays local on your machine. Nothing is sent anywhere except public GitHub API calls to fetch your own repos.

Related MCP server: ResumePilot MCP

Setup

1. Install the dependency

cd skillmatch-mcp
pip install -r requirements.txt

Speaks MCP protocol revision 2026-07-28 over stdio, on SDK mcp>=2.0,<3. It is stateless: no initialize handshake and no session id, with every request carrying its own protocol version and client identity. tools/list and server/discover return ttlMs/cacheScope freshness hints, and tool order is stable across calls so client-side and prompt caches keep hitting.

Upgrading from 1.x: 2.0.0 drops the initialize handshake, so it requires a client that speaks 2026-07-28. Tool names, schemas, and results are unchanged — only the transport moved. Stay on 1.x if you need to serve an older client.

2. Add to Claude

Claude Code (CLI):

claude mcp add --scope user skillmatch -- python /path/to/skillmatch-mcp/server.py

Then restart Claude Code.

Claude Desktop:

Open your Claude Desktop configuration file and add SkillMatch to the mcpServers section:

{
  "mcpServers": {
    "skillmatch": {
      "command": "python",
      "args": ["/path/to/skillmatch-mcp/server.py"]
    }
  }
}

Replace /path/to/skillmatch-mcp/server.py with the actual path to server.py on your machine, then restart Claude Desktop.

3. Start using it

After restarting, SkillMatch tools will be available in your conversation. Works in Claude Code and Claude Desktop.

Onboarding

The first time you use SkillMatch, Claude will walk you through a setup conversation. It looks something like this:

You: Help me find a new job.

Claude: I don't have your profile set up yet. Let me ask you a few questions first.

  What is your name?
You: Jordan

Claude: What is your current role?
You: Senior frontend developer

Claude: What roles are you targeting?
You: Staff frontend engineer, senior full-stack engineer

Claude: What is your minimum acceptable salary?
You: 180000

Claude: Are you only looking for remote positions?
You: Yes

Claude: Where are you located?
You: Portland, OR

Claude: Any dealbreakers?
You: No on-call rotations, no defense contractors

Claude: What is your GitHub URL?
You: https://github.com/jordan-dev

Claude: Where is your resume file?
You: C:/Users/jordan/Documents/resume.md

Claude: Got it. Your profile is saved. Let me search for jobs matching your targets.

After onboarding, your profile is saved locally and reused in every future conversation.

Extended Profile (optional)

You can enrich your profile at any time by asking Claude to update it. These fields help Claude reason about nuanced fit:

Field

Type

Example

work_style

object

{async_preferred: true, ic_vs_leadership: "ic", client_facing_tolerance: "none", team_size_preference: "small"}

optimizing_for

array

["comp", "autonomy", "interesting_problems"]

unlisted_skills

array

["MCP protocol implementation", "Windows installer packaging"]

developing_skills

array

["TypeScript", "RAG pipelines"]

dealbreaker_detail

array

[{dealbreaker: "on-call", hardness: "absolute"}, {dealbreaker: "relocation", hardness: "strong_preference"}]

rejection_patterns

array

["roles that sounded like automation but were actually IT support"]

These can be set during initial setup or added later with the update_profile tool.

How It Works

search_jobs builds a search query from your profile and any keywords you provide. Claude then uses that query with its web search capabilities to find real listings. The tool itself does not search the web.

analyze_fit runs a two-step process. First it parses the job description into structured signal (hard requirements, nice-to-haves, red flags, compensation signals, role type) via the Claude API. Then it fetches your portfolio and auto-selects the best resume variant for the detected role type. Claude sees structured signal before raw marketing copy.

parse_jd is the standalone JD parser. Use it independently to pre-process a job description without running the full fit analysis.

log_application, get_applications, and update_application form a job search CRM. Track status (applied, screening, interview, offer, rejected, ghosted), set follow-up dates, and record outcomes.

get_follow_ups shows applications that need attention — where the follow-up date has passed and you're still waiting.

get_application_patterns analyzes your full application history (10+ needed) to find which role types get responses, which skills resonate, and recommends search adjustments.

email_ranked_jobs.py is a standalone Conductor worker script that emails the latest ranked job report. See Email Worker Setup below for configuration.

save_scouted_job saves a job listing found during scouting. It validates the URL to reject search result pages and deduplicates against existing scouted jobs and applications.

get_scouted_jobs returns all scouted listings, optionally filtered to only unranked ones. mark_jobs_ranked marks all unranked jobs as ranked after a ranking report is generated.

add_resume and list_resumes manage multiple resume variants. Each variant targets specific role types (e.g. "AI Engineering" targets ai_engineering and ml_engineering). During fit analysis, the best variant is auto-selected based on the JD's detected role type.

update_profile merges new or changed fields into your existing profile without re-running setup.

get_portfolio and get_resume can be called independently if you want Claude to review just your repos or just your resume.

Quick Start (No File Paths)

For the simplest setup, paste your resume directly — no local files needed:

You: Help me find a job.
Claude: What is your name?
You: Alex
Claude: What roles are you targeting?
You: AI engineer, ML engineer
Claude: Paste your resume or provide a file path.
You: [paste resume text here]
Claude: Profile saved. Let me search for jobs.

Email Worker Setup

email_ranked_jobs.py sends the ranked job report via Gmail. It is designed to run as a Conductor worker on a schedule.

Credentials live in a gitignored config file — never in the source. Copy the template and fill it in:

cp data/email_config.example.json data/email_config.json

Then edit data/email_config.json:

{
  "gmail_user": "you@gmail.com",
  "gmail_app_password": "your-16-char-app-password",
  "email_to": "you@gmail.com",
  "email_cap": 15
}

Key

Required

Default

gmail_user

yes

—

gmail_app_password

yes

—

email_to

no

falls back to gmail_user

email_cap

no

15

The whole data/ folder is gitignored (only .gitkeep and email_config.example.json are tracked), so email_config.json can never be committed by accident.

Why a config file and not environment variables? Depending on how the worker process is launched — notably via Conductor on Windows — environment variables may not be inherited by the child process, causing silent send failures. The config file is read from a path relative to the script, so it works regardless of how the process was spawned. Environment variables (GMAIL_USER, GMAIL_APP_PASSWORD, EMAIL_TO, EMAIL_CAP) are still honored as a fallback when a key is absent from the config file, which is handy for CI or containerized runs. If neither source supplies credentials, the script exits with code 2 and prints setup instructions instead of failing silently.

To generate a Gmail App Password: Google Account → Security → 2-Step Verification → App Passwords. Create one for "Mail". This is not your normal account password. If you ever paste one into a source file by mistake, revoke it from that same screen.

The script reads data/ranked_jobs.md, caps the email to the top 15 ranked listings, sends it, and then deletes ranked_jobs.md so stale rankings are not recycled on the next run.

File Structure

skillmatch-mcp/
  server.py              # MCP server (stdio JSON-RPC)
  requirements.txt       # python-docx dependency
  CLAUDE.md              # Instructions for Claude
  README.md              # This file
  email_ranked_jobs.py   # Conductor worker: emails ranked job reports
  cowork_monitor.py      # Conductor worker: monitors Cowork VM, auto-recovers
  cowork_tab.png         # Reference image for Cowork tab UI automation
  data/                  # Entire folder gitignored except the two tracked files below
    .gitkeep                  # Keeps the folder in git (tracked)
    email_config.example.json # Credential template (tracked, no real secrets)
    email_config.json         # Your real Gmail credentials (gitignored)
    profile.json              # Created on first setup (gitignored)
    applications.db           # Created on first log (gitignored)
    scouted_jobs.json         # Scouted listings (gitignored)
    ranked_jobs.md            # Latest ranked report (gitignored)

Available Tools

20 tools
add_resumeB

Add a new resume variant to the profile. Each variant targets specific role types for automatic selection during fit analysis.

Inputs: id (required), label (required), role_types (required), path, +1 more.

analyze_fitA

Gather all user data needed for job fit analysis. Accepts a job description, then internally fetches the user's portfolio (GitHub repos) and resume. Returns all three together in a structured bundle so Claude can reason about the fit. Does NOT perform the analysis itself.

Inputs: job_description (required).

get_application_patternsA

After 10+ applications, analyze the full history to find patterns: which role types get responses, which skills resonate, which red flags recur in silent roles, and recommended search adjustments. Uses Claude API internally.

Inputs: no named parameters.

get_applicationsB

Return all tracked job applications, ordered by most recent first. Optionally filter by status.

Inputs: status.

get_follow_upsA

Return all applications where follow_up_due_date is today or earlier and status is still applied or screening. Shows what needs attention.

Inputs: no named parameters.

get_portfolioA

Fetch public GitHub repos for the user. Returns repo name, description, language, topics, last updated, homepage, and README content for each repo. Pass deep=true to also run a Haiku-powered analysis of each README, extracting stack, what it does, complexity, and notable achievements. Omit deep (or pass false) for the free metadata+README fetch. Only use deep=true when you explicitly need structured analysis and the user has authorized API spend.

Inputs: deep.

get_profileA

Read and return the saved user profile from data/profile.json.

Inputs: no named parameters.

get_resumeA

Read the user's resume file. Reads resume_path from the saved profile and returns the raw text content. Supports .txt, .md, and .docx files.

Inputs: no named parameters.

get_scouted_jobsB

Return all scouted job listings from scouted_jobs.json. Optionally filter to only unranked jobs.

Inputs: unranked_only.

list_resumesA

Return all stored resume variants with their labels and target role types.

Inputs: no named parameters.

log_applicationB

Log a job application to the tracking database. Creates the database if it does not exist.

Inputs: role (required), company (required), url, notes, +2 more.

mark_jobs_rankedA

Mark all unranked scouted jobs as ranked. Returns the count of jobs marked.

Inputs: no named parameters.

parse_jdA

Parse a job description into structured signal: hard requirements, nice-to-haves, responsibilities, red flags, compensation signals, role type, experience level, and domain. Uses Claude API internally. Returns structured JSON for better fit reasoning.

Inputs: job_description (required).

purge_dead_listingsA

Checks every scouted job URL for liveness and removes expired or unavailable postings from scouted_jobs.json. Detects dead listings via HTTP 404/410 status codes and known closure phrases in page content (e.g. 'no longer accepting applications'). Connection errors and 5xx responses are treated as 'keep' to avoid false positives. Pass dry_run=true to preview what would be removed without actually deleting anything.

Inputs: dry_run.

read_repoA

Analyzes a local code repository and extracts structural information: detected languages, tech stack (frameworks/libraries from imports), entry points, key modules (classes and functions), and architecture shape. Returns raw structured JSON for Claude to interpret — no API calls, pure local analysis using AST for Python files and regex for JS/TS. Use this to understand a candidate's project before analyzing job fit, or to supplement get_portfolio with deeper technical signal from local repos.

Inputs: path (required).

save_scouted_jobA

Save a scouted job listing to the tracked scouted_jobs.json file. Validates the URL to ensure it is a DIRECT link to a specific job posting (e.g. indeed.com/viewjob?jk=..., greenhouse.io/.../jobs/...) and rejects search result page URLs (e.g. indeed.com/q-..., builtin.com/jobs/...). Also deduplicates against existing scouted jobs and applications. Returns {saved: true/false, reason: '...'} with clear feedback.

Inputs: url (required), role (required), company (required), remote, +3 more.

search_jobsA

Build an optimized job search query. Accepts a free-text query and combines it with the user's saved profile (target roles, salary floor, remote preference) to construct a search query string. Returns the query for Claude to use with web search tools. Does NOT perform the search itself.

Inputs: query (required).

setupB

Onboard the user by collecting their job search profile. Captures: name, current_role, target_roles (list), salary_floor (integer), remote_only (bool), location, dealbreakers (list), github_url, resume_path. If a profile already exists, the 'confirm_overwrite' parameter must be true to replace it.

Inputs: name (required), location (required), github_url (required), remote_only (required), +14 more.

update_applicationB

Update an existing application by ID. Accepts any subset of fields: status, notes, follow_up_due_date, response_received, outcome. Automatically updates last_activity_date.

Inputs: id (required), notes, status, outcome, +2 more.

update_profileA

Update the user's profile with new or changed fields. Accepts any subset of profile fields and merges them into the existing profile.json without requiring a full re-setup. Useful for adding unlisted_skills, updating dealbreaker_detail, or changing any field.

Inputs: name, location, github_url, work_style, +11 more.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.9/5.0

    Scored across 20 tools

    Disambiguation5/5

    Each tool serves a distinct purpose with clear boundaries: profile management, application tracking, resume handling, portfolio analysis, job scouting, and utility functions. No two tools overlap significantly; descriptions are detailed enough to prevent confusion.

    Naming Consistency5/5

    All tool names follow a consistent verb_noun pattern in snake_case, e.g., get_profile, search_jobs, update_application. The only slight deviation is 'setup', but it still reads as a clear action. No mixing of conventions.

    Tool Count4/5

    With 20 tools, the server is on the higher side of reasonable for a comprehensive job search assistant. The count covers profile, resume, portfolio, application, and scouting workflows without being excessive. Slightly over the ideal 3-15 range but still well-scoped.

    Completeness4/5

    The tool set covers most CRUD operations for key entities: profile, applications, scouted jobs, and resumes. Minor gaps like explicit delete operations for applications or resume variants exist, but agents can work around them using update or other means. The domain is well-covered.

    Maintenance

    ActivityInactive
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • F
      license
      Not graded
      quality
      D
      maintenance
      Enables Claude to parse CVs, search job boards (Remotive, Arbeitnow, Adzuna, Greenhouse/Lever), tailor resumes and cover letters, and prepare application packages with direct apply links—without ever auto-submitting. It runs 100% locally and free, storing jobs and applications as JSON files.
      -
    • A
      license
      A
      quality
      C
      maintenance
      Enables parsing a master resume, analyzing job descriptions for ATS/match scores, generating truthful tailored resumes, tracking applications, and producing PDF reports through Claude Desktop.
      17
      MIT
    • F
      license
      Not graded
      quality
      B
      maintenance
      Enables automated job searching from multiple boards, filtering matches against saved criteria, deduplicating seen postings, and tracking application statuses. Runs locally and integrates with Claude Desktop for natural-language queries and updates.
      -
    • A
      license
      B
      quality
      B
      maintenance
      Enables Claude to search job listings on selected sites, filter and score them against your own skills, track application status, and generate a tailored LaTeX/PDF resume for each job. It also supports importing an existing Word or PDF resume and diagnosing local setup, all with data kept in a private folder on your own machine.
      12
      MIT