runwayMCP

An MCP server that helps international students (F-1/OPT) filter US job postings by technical fit AND visa sponsorship history — in a single call.

Quick install

Option A: Claude Code plugin (recommended — two commands)

/plugin marketplace add satovarb16/runwayMCP
/plugin install runway-mcp@satovarb

Claude Code wires up the MCP server for you — no JSON to edit.

Updating: new releases arrive through the normal plugin flow — no need to touch PyPI.

/plugin marketplace update satovarb
/plugin update runway-mcp@satovarb

Then run /reload-plugins (or restart Claude Code) to load the new version. The plugin pins an exact package version, so updating it pulls the matching server release.

Option B: manual .mcp.json

Create a .mcp.json file in the directory where you run Claude Code:

{
  "mcpServers": {
    "runway-mcp": {
      "command": "uvx",
      "args": ["runway-mcp"]
    }
  }
}

That's it. Open Claude Code — uvx downloads and runs the server automatically.

Don't have uv? Install it: pip install uv (or see uv docs)

Alternative: install from source

git clone https://github.com/satovarb16/runwayMCP
cd runwayMCP
pip install -e ".[dev]"

Then use python -m server instead of uvx runway-mcp in your .mcp.json, and add "cwd": "/path/to/runwayMCP".

Optional extra: parsing Greenhouse custom domains needs Playwright. Most users can skip it — see Optional: Playwright.

Related MCP server: h1b-mcp

Step 0 (required): ingest your CV

Do this once before anything else. analyze_job needs a stored profile — without it it returns an error asking you to run this first.

You: "Set up my profile using my CV at /path/to/resume.pdf"

Accepted CV formats: .pdf and .docx only.

Claude reads your CV, extracts a structured profile, and saves it locally at ~/.config/runway-mcp/profile.json. Updated your CV later? Just say "Update my profile with my new CV at ..." to replace it.

Usage

You: "Evaluate this role for me: https://jobs.example.com/swe-123"
Claude:
  → analyze_job(url) — fetches job + checks visa + loads your profile
  → scores the CV match and returns APPLY / CONSIDER / SKIP + reasoning

On first run, the server downloads USCIS H-1B data (~2MB) automatically.

Optional: Playwright for JavaScript-heavy job boards

You almost certainly don't need this. It's only for parsing Greenhouse custom domains (a rare edge case). Canonical boards.greenhouse.io, Ashby, and Lever URLs always work without it. The server prints a harmless warning at startup if Playwright is missing — you can ignore it unless you hit a custom-domain Greenhouse URL.

Because uvx runs the server in an isolated environment, installing Playwright globally won't reach it — you must pull in the browser extra so it lands in the server's env.

If you installed via uvx / the plugin, switch to a manual .mcp.json that requests the extra:

{
  "mcpServers": {
    "runway-mcp": {
      "command": "uvx",
      "args": ["--from", "runway-mcp[browser]", "runway-mcp"]
    }
  }
}

If you installed from source:

pip install -e ".[browser]"

Then, either way, download the browser binary once:

playwright install chromium

How it works

Claude Code launches this server over stdio and calls its tools when relevant. You don't invoke the tools directly — Claude decides when to call them based on the conversation.

The tools fetch and shape data; Claude does the reasoning. The server never calls back to the model (no MCP sampling), so it works on any MCP host — including Claude Code, which does not support sampling. Claude extracts your profile from the CV and scores the job-vs-profile match itself, using the rubric the tools return.

One-call flow (recommended):

You: "Evaluate this role for me: https://jobs.example.com/swe-123"
Claude:
  1. analyze_job(url) → job details + visa verdict + your profile + scoring guide
  2. [scores the match + applies the rubric] → APPLY/CONSIDER/SKIP, red flags, advice

Or use the individual tools directly:

Claude:
  1. fetch_job_posting(url)          → job title, company, country, full JD
  2. check_visa_sponsorship(company) → H-1B history, approval rate, verdict
  3. get_profile()                   → your stored CV, to score the fit against

The visa check only runs for US roles — Claude skips it for positions in other countries.

Status

Tools

analyze_job(url: str) -> AnalyzeJobResult

One-call data gatherer. Fetches the job, checks visa sponsorship, and loads your stored profile, then returns a combined envelope plus a scoring guide. Claude scores the match and applies the recommendation rules — the server does not (no MCP sampling). Returns:

{
  "job":     { "title": "...", "company": "...", "url": "..." },
  "visa":    { "verdict": "GREEN", "filings": 42, "approval_rate": 0.91 },
  "profile": { "name": "...", "skills": [...], "experience": [...] },
  "scoring_guide": {
    "instructions": "Score the match 0-100 and apply the rules...",
    "recommendation_rules": ["SKIP if visa RED or score < 40 ...", "..."]
  }
}

Recommendation thresholds (Claude applies these from the scoring guide):

• APPLY — visa GREEN and score ≥ 70

• SKIP — visa RED or score < 40 (SKIP takes precedence)

• CONSIDER — everything else

Requires a stored profile (run setup_profile first). If no profile exists, returns a clear error.

setup_profile(profile: ProfileData) -> ProfileSetupResult

Persists a structured profile to ~/.config/runway-mcp/profile.json. Claude reads your CV (.pdf or .docx) and extracts the ProfileData (name, skills, experience, education, …), then calls this tool to save it. Fails if a profile already exists — use update_profile to replace it. Required before analyze_job.

update_profile(profile: ProfileData) -> ProfileSetupResult

Same as setup_profile but overwrites an existing profile. Use when you update your CV.

get_profile() -> GetProfileResult

Returns the stored profile so Claude can score a job against it (e.g. in the individual-tools flow). Returns a structured no_profile error if none is stored yet.

check_visa_sponsorship(company: str) -> VisaResult

Looks up a company's H-1B petition history via the USCIS H-1B Employer Data Hub.

Returns: company, total_filings, approval_rate (0–1), verdict (green/yellow/red), source.

Verdict thresholds (calibrated against FY2024 data, ~36k employers):

• green — ≥ 5 filings AND approval rate ≥ 80% (active sponsor, top ~10%)

• yellow — ≥ 1 filing AND approval rate ≥ 50% (has sponsored before)

• red — no record or rate below threshold

Data is downloaded and cached at ~/.cache/runway-mcp/uscis_h1b.csv on first call (~2MB) and auto-refreshes to the latest FY on every server startup.

fetch_job_posting(url: str) -> JobPostingResult

Fetches and parses a job posting from a URL.

Returns: title, company, country, location, description, posted_date, source_url.

Supported job boards

Known gaps

Tool vs. reasoning boundary

These tools only fetch and shape data. Claude handles all reasoning:

• Whether to call check_visa_sponsorship (only for US roles)

• How to interpret the verdict and score in context

• Whether the role is a good fit overall

This is intentional — tools that encode judgment make Claude less useful, not more.

Tests

pytest -m contract      # fast contract tests
pytest -m integration   # server tool registration
pytest                  # full suite (214 tests)

Contributing

pip install -e ".[dev]"
pre-commit install       # runs ruff lint + format before every commit

Highest-value next features (in priority order):

1. Workday parser — dedicated parser for better reliability on Workday boards

2. SmartRecruiters — public API, clean integration

3. BambooHR — public API, clean integration

4. Lever custom domains — same pattern as Greenhouse custom domains

PRs welcome.

License

MIT © satovarb