roasted.cv mcp Connect your agent

hosted MCP server · mcp.roasted.cv

Your agent runs the job hunt. You show up to interviews.

roasted mcp hands any MCP client the entire roasted.cv job hunt as 23 tools: discover fresh roles, score fit against your resume, tailor every application, render a one-page PDF, and track the pipeline. A resume is a ticket. Interviews are the point.

$ claude mcp add --transport http roasted https://mcp.roasted.cv/mcp

One-time OAuth sign-in, no API keys. Using something other than Claude Code? Pick your client below.

The entire toolkit

You prompt. The agent grinds.
Interviews land on your calendar.

Six moves, one chat window. The numbers below are fictional; the workflow is exactly this.

A ranked shortlist, scored against your resume

match_jobs

Find senior TypeScript roles in Berlin that actually fit my resume.

match_jobs

Searched 46 fresh postings as a guest, no login. Pulled the full JD for the top 8 and scored each against your resume. Three are worth your time:

match_jobs · ranked by fit
  • 92fit

    Senior TypeScript Engineer

    Aurelia Systems · Berlin · Hybrid · 2d ago

    TypeScript React Node.js GraphQL Kubernetes

  • 87fit

    Staff Frontend Engineer

    Kupferwerk · Berlin · On-site · 1d ago

    TypeScript React Next.js Vue

  • 74fit

    Senior Full-Stack Engineer

    Nordwind Labs · Remote, EU · 3d ago

    TypeScript Node.js Go Terraform

Every application tailored, never from a template

apply_create get_kit

Tailor an application for the Aurelia role.

apply_create

Kit ready. Resume re-cut for their stack with JD keywords woven in, cover letter kept short and formal. Every claim in it comes from your resume, nothing invented.

get_kit · kit_9f2c41

status: ready38s

  • tailored resume

    re-cut for Aurelia's stack · intensity: bold

  • cover letter

    96 words · professional · formal

  • JD signals

    techStack: 12 · keyRequirements: 5

A one-page PDF, every time

render_pdf

Give me the PDF.

render_pdf

Rendered. The first pass spilled onto a second page, so it auto-condensed and re-rendered. One page, recruiter-ready.

render_pdf · Lena-Hartmann-Resume.pdf

pageCount: 1

shrink: true

spilled past one page, condensed 4 bullets, re-rendered

Handoff with the exact apply URL

resolve_apply_url apply_package

Where do I actually apply?

apply_package

On Aurelia's own site. Here is the exact apply URL the posting records, bundled with your PDF and cover letter. You open it and hit submit.

apply_package · handoff

exact company apply URLapplyType: offsite

job-boards.greenhouse.io/aurelia/jobs/4829131

  • Lena-Hartmann-Resume.pdf

    one page · neutral file name

  • cover-letter.txt

    96 words · ready to paste

You submit on the company site. Applications never go through LinkedIn.

Screening questions, answered from your facts

answer_form

The form asks: "Why Aurelia?" and "How many years of production React experience do you have?"

answer_form

Answers below, grounded in your resume and nothing else. Paste them in as they are.

answer_form · 2 questions
Why do you want to work at Aurelia?
Aurelia is rebuilding checkout for European retail, which is the shape of problem I have spent six years on: typed APIs under real traffic and React frontends a team can actually maintain.
How many years of production React experience do you have?
Six years in production, from a component library used by 40 engineers to consumer checkout flows.

grounded in your resume, never improvised

A pipeline you can query, not a spreadsheet

track_stats

How is the pipeline looking this month?

track_stats

38 applications out, 6 interviews on the calendar. Response rate is 21 percent and climbing. Want me to queue the next batch?

track_stats · last 30 days

38

applied

6

interviewing

12

saved

9

closed

response rate 21% · interview rate 16%

All 23 tools

The entire job hunt, all in one server.

Onboarding to offer. One OAuth sign-in, no API keys, nothing to install on your machine.

  • roasted_onboarding
  • save_candidate_profile
  • get_profile
  • account_info
  • list_cvs
  • get_cv
  • create_cv
  • update_cv
  • discover_jobs
  • match_jobs
  • get_jd
  • resolve_apply_url
  • answer_form
  • apply_create
  • get_kit
  • render_pdf
  • get_cover_letter
  • apply_package
  • track_add
  • track_list
  • track_get
  • track_status
  • track_stats

Portable playbook

One SKILL.md carries the judgement.

Workflow, hard rules, and honest-application guardrails live in a single Agent Skill file, so the loop runs the same way in Claude Code, Codex, and Cursor. Connect the server, drop in the skill, and your agent already knows how to hunt.

Connect

One server. Every client.

The same hosted server at https://mcp.roasted.cv/mcp connects everywhere. Pick your client for the exact snippet, then authorize once. The server name is roasted, no hyphen.

FAQ

Fair questions.

How does my agent connect?

The server speaks MCP over streamable HTTP at https://mcp.roasted.cv/mcp with OAuth 2.1. No API keys: your client opens a browser tab once, you sign in to your roasted account, and every session after that is automatic. Clients that cannot open a browser, like Continue.dev and JetBrains AI Assistant, use a personal access token from the roasted dashboard instead.

Which agents does it work with?

Anything that speaks MCP: Claude Code, Claude Desktop, Codex, Cursor, VS Code, Windsurf, Zed, Continue.dev, and JetBrains AI Assistant all connect to the same hosted server. The server name is roasted and the URL never changes.

What is free and what needs a paid plan?

Discovery, matching, full job descriptions, resume authoring and editing, account info, and the whole application tracker are free. Three tools need a paid roasted plan: apply_create for tailoring, resolve_apply_url for the exact company apply link, and answer_form for grounded screening answers.

Is this safe for my LinkedIn account?

Yes. The server never logs into LinkedIn and never asks you to. Job discovery and descriptions are read as a guest, and every application is submitted by you on the company's own site, never through LinkedIn.

What exactly is in a kit?

A resume tailored to that one job description, a short formal cover letter grounded in your real experience, and the parsed JD signals used for tailoring. render_pdf turns the resume into a one-page PDF, and apply_package bundles everything with the exact apply URL for handoff.

What if a call times out and I retry?

Nothing bad. apply_create is idempotent: retrying with the same resume and job description within about ten minutes returns the same kit and never charges you twice.