job-offer-scraper-mcp exposes a read-only MCP tool that turns public job URLs into structured descriptions and criteria. This repository also ships the remotely installable tailor-latex-cv Codex skill, which adapts a real LaTeX CV to an offer, compiles it with a verified Tectonic binary, and checks that the PDF remains machine-readable.

✨ Highlights

🚀 Quick start

Run the published package in an isolated UV environment:

uvx job-offer-scraper-mcp

Or run directly from GitHub:

uvx --from git+https://github.com/JuanjoLopez19/job-scrapper-mcp.git job-offer-scraper-mcp

MCP client configuration

{
	"servers": {
		"job-offer-scraper": {
			"command": "uvx",
			"args": ["job-offer-scraper-mcp"]
		}
	}
}

To use the GitHub source before or instead of a PyPI release:

{
	"servers": {
		"job-offer-scraper": {
			"command": "uvx",
			"args": [
				"--from",
				"git+https://github.com/JuanjoLopez19/job-scrapper-mcp.git",
				"job-offer-scraper-mcp"
			]
		}
	}
}

🌐 Supported job sites

The MCP exposes:

get_job_offer_details(url: string)

Successful responses contain description and criteria, or generic page content. Failures use a structured error object with a stable code.

🧩 Install the LaTeX CV tailoring skill

Ask Codex to install the skill from this GitHub path:

https://github.com/JuanjoLopez19/job-scrapper-mcp/tree/master/skills/tailor-latex-cv

The skill and MCP server are separate components: install the skill from GitHub, then register the MCP server with uvx using the configuration above.

Workflow

flowchart LR
    A[LaTeX CV + job URL] --> B{Job Offer MCP}
    B -->|Success| E[Evidence-based match]
    B -->|Unavailable| C[Web fallback]
    C -->|Unavailable| D[Ask user for offer text]
    C --> E
    D --> E
    E --> F[Truthful visible tailoring]
    F --> G[Tectonic compilation]
    G --> H[pypdf text verification]
    H --> I[Atomic PDF publication]
    I --> J[Tailored .tex + verified PDF]

The workflow preserves the original CV, integrates only supported keywords in visible recruiter-readable text, and reports material requirements that the CV does not substantiate.

🧪 Compile and verify a tailored CV

The skill includes a portable Python utility managed by UV:

uv run skills/tailor-latex-cv/scripts/compile_latex.py cv-tailored.tex \
  --install-tectonic \
  --expected-keyword Python \
  --expected-keyword "REST APIs"

On first use, --install-tectonic downloads the pinned official Tectonic 0.16.9 binary for the current platform and verifies its SHA-256 digest. Compilation runs in Tectonic's untrusted mode using a pinned direct official resource bundle URL. The generated PDF is then opened with pypdf, which verifies page count, extractable text, and requested keywords. Only after those checks pass, the utility atomically publishes the final PDF beside the tailored .tex.

Omit --install-tectonic to prohibit compiler downloads. Use --forbidden-keyword to fail verification if an unsupported requirement appears in the generated PDF. Use --final-pdf <path> to select another deliverable location. Existing PDFs are preserved with a numbered suffix unless --overwrite-final is explicitly supplied.

[!NOTE] On Windows, replace decorative fontawesome5 icons with visible contact labels in the tailored copy. The verifier detects this package before compilation because the pinned Windows Tectonic build cannot load it reliably.

🛡️ Integrity and security

• The MCP accepts only public HTTP(S) targets and blocks unsafe local resources.
• The MCP tool is annotated as read-only, idempotent, and non-destructive.
• Job-page content is treated as untrusted input rather than agent instructions.
• CV tailoring never invents experience or inserts invisible ATS keywords.
• Tectonic release archives are pinned and hash-verified before execution.
• LaTeX compilation uses --untrusted to disable known-insecure engine features.

🛠️ Development

Requirements:

• Python 3.11 or newer
• UV

Install dependencies and enable the quality hooks:

uv sync
uv run pre-commit install

Run the complete quality suite:

uv run pre-commit run --all-files

Run the MCP from the working tree:

uv run job-offer-scraper-mcp

Build and smoke-test the distribution:

uv build --no-sources
uv run --isolated --no-project --with dist/*.whl tests/smoke_test.py

📁 Project layout

├── src/job_offer_scraper_mcp/   # MCP server and scraper implementations
├── skills/tailor-latex-cv/      # Remotely installable Codex skill
│   ├── agents/openai.yaml
│   ├── references/
│   ├── scripts/compile_latex.py
│   └── SKILL.md
├── tests/                       # Unit, integration, and smoke tests
├── pyproject.toml               # UV project and quality configuration
└── .pre-commit-config.yaml      # Ruff, Pyrefly, and pytest hooks

📦 Publishing

Tags beginning with v trigger .github/workflows/release.yml, which builds, smoke-tests, and publishes the package to PyPI through Trusted Publishing.

📄 License

Distributed under the MIT License.