Crosswalk

An AI-native career copilot that lives inside your AI, not on a website. Local-first. Zero API keys. Bring your own model.

Crosswalk is an MCP server. Install it once into Claude Desktop (or any MCP host that supports sampling). Your AI gains 6 tools to find roles, score fit, and tailor applications — using the model you already pay for.

Quick start

npx crosswalk-mcp install

Restart Claude Desktop. Then say:

"Set up my profile: I'm a PM with 2 years at Acme, want NYC or remote, looking at AI infra." "Add my resume from ~/Documents/resume.pdf, label it 'Generic PM'." "Find PM roles at H-1B sponsors with 0.8+ confidence." "Why am I a fit for the Stripe Payments PM role?"

What it does (M3)

Tool	Purpose
setup_profile	Store a structured profile from a free-form description.
add_resume	Parse and store a labeled resume version (DOCX/PDF/text).
list_resumes	List stored resumes.
fetch_jobs	Pull live roles from Greenhouse, Lever, and Ashby.
score_fit	Numeric fit score + structured strengths/gaps.
explain_fit	Markdown narrative — why fit, gap, positioning.
tailor_resume	Edit your best base resume for a specific JD; returns markdown, DOCX, or print-ready HTML.
draft_application	Build a complete application bundle (tailored resume + cover letter + deep link), persisted as a tracked draft. Anti-spam guardrail enforces a weekly cap and refuses obvious duplicates.
submit_application	Mark an application submitted after you click Apply in your browser.
set_status	Update an application status (interviewing, rejected, offer, etc.).
add_note	Append a free-text note to an application's event log.
list_pipeline	List your applications with company + job context, optionally filtered by status.
schedule_workflow	Schedule a non-sampling recurring workflow (job-cache refresh, old-job pruning) via cron expression.
run_workflow	Manually run a scheduled workflow now.
list_workflows	List all scheduled workflows.
delete_workflow	Delete a scheduled workflow by id.

Why it's different

1. Zero API keys. All AI work runs through MCP sampling — calling back into your AI client's model. No keys in this repo, no AI bill on us, no rate limits beyond yours.
2. Local-first. Profile, resumes, and job cache live in ~/.crosswalk/state.db. Your data never leaves your machine.
3. Open Job Graph. The company → ATS registry is a checked-in, MIT-licensed JSON dataset. Add your favorite companies via PR.

Roadmap

Version	Headline
M1	Discover + match + explain
M2	Tailor resume, draft cover letter, application "PR" bundle
M3 (this release)	Pipeline tracker, anti-spam guardrail, scheduled workflows
M4	7 more ATS adapters; registry to 200+ companies; install polish
v2	Autonomous apply via Playwright in a sandbox

See docs/superpowers/specs/2026-04-30-crosswalk-design.md for the full spec.

Scheduled workflows (optional)

Crosswalk can run non-sampling workflows (job cache refresh, old-job pruning) on a schedule. These don't need the AI host to be running — Crosswalk pokes the ATS APIs directly.

Schedule one in chat:

"Schedule a workflow to refresh PM jobs at H-1B sponsors every Monday at 9 AM."

Then add a single line to your crontab to actually invoke them:

* * * * * /usr/local/bin/crosswalk-mcp run-scheduled >> ~/.crosswalk/scheduler.log 2>&1

Sampling-driven workflows (e.g., "tailor the top 3 fits") are a v2 feature — they need a live AI host.

Development

npm install
npm test           # run vitest
npm run dev        # run the MCP server over stdio
npm run build      # bundle dist/server.js + dist/cli.js

License

MIT.