Sign in

MCP server / Documentation

Career copilot skill

A guided job-search coach for your AI assistant, with nothing to install, that checks how well you fit a real role, writes an honest ATS-friendly resume from your real projects, and tells you what to build next when you are not ready yet.

Back to the MCP page

Overview

Career copilot is one skill with three parts that work on their own or in order:

  • Job match reads a role with the MCP server and scores how well your profile and GitHub projects fit it, with the evidence for every point.
  • ATS resume writes a clean, one-column resume that applicant tracking systems parse correctly, tailored to the role and using only what you can show.
  • Suggestions turns gaps into a concrete plan: what to build, in what order, and which roles fit you today.

It is built around one rule: only real facts. If you do not meet a role yet, it tells you so and shows the way, instead of padding your resume.

What you need

  • An AI assistant that supports MCP servers (for example Claude, ChatGPT or Cursor) and lets you start an MCP prompt.
  • The JoinAStartUp MCP server connected. The skill uses its get_job, search_jobs, get_company, search_companies and score_fit tools. The MCP server and the Career copilot that comes with it are Premium features.
  • Your details and, ideally, links to your GitHub projects. Public repos work best. For a private repo, paste the README and the file list.

How to use (no install)

There is nothing to download or install. The whole skill is served by the MCP server, so it always matches the latest version and never takes space on your computer.

  1. Connect the MCP server to your assistant, following the MCP page, and sign in with a Premium account.
  2. In your assistant, open its list of MCP prompts or commands and choose career_copilot. Where that list appears depends on the assistant.
  3. Optionally fill in task (match, resume, suggestions or all) and job (a job link or slug from this site). Leave them empty to be guided.
  4. Answer the first questions. The assistant builds your profile, reads the job, and carries on from there.

If your assistant does not list MCP prompts, you can still ask it in plain words to find roles, read a job and check your fit, because the job tools work either way. Only the full guided flow, with the profile, the verdict rules and the resume files, comes from the prompt.

Premium and your access

Career copilot is included with Premium. Without Premium you can read this page, but your assistant cannot connect to the MCP server or start the prompt: you will see Upgrade to use this with a link to upgrade.

If your Premium ends while you are connected, access stops at your assistant's next request. Nothing has to expire first, because Premium is checked on every call. Your assistant then shows an upgrade message, and it cannot look up more jobs or run the fit score. What is already in that conversation stays there. After you upgrade, reconnect from your assistant and carry on.

Premium ends when a cancelled subscription reaches the end of the period you paid for, when a payment fails, after a refund or a lost dispute, or on the end date of a membership granted by us. Cancelling keeps Premium until the end of the paid period.

Quick start

Start a new conversation and choose the career_copilot prompt, or ask in plain language. Paste a job link from the site and, if you have one, your GitHub link or resume.

  • “Do I fit this role? Here is the link, and my GitHub is github.com/asha.”
  • “Make my resume ATS friendly for this job. Here is my current resume.”
  • “Which of the remote frontend roles on joinastartup suit me best?”
  • “I am not ready for this role. What should I build?”

The first time, the assistant asks for your details and keeps them as a JSON profile. It can save that to a file so the next role takes seconds.

How it works

  1. Profile. It gathers your name, contact, education, experience, projects and certifications, then learns each project from its GitHub repo: README, dependencies, Dockerfile, CI, tests and structure.
  2. Job match. It reads the role with get_job, lists the must-haves and nice-to-haves, and checks each against your evidence.
  3. Next step by verdict. Strong goes to the resume. Partial writes the resume without the gaps and adds a plan. Not yet skips the resume and gives you the plan, plus roles that fit today.

A skill counts as shown only when the repo or you prove it. A Dockerfile in the repo shows Docker. The word “scalable” in a README does not show load balancing.

Fit score and verdicts

Each requirement is shown (1 point), partly shown (half a point) or missing (0). The score is 80% must-haves and 20% nice-to-haves, out of 100. A missing blocker, such as required years of experience, a licence or work authorisation, caps the verdict.

Your assistant judges each requirement, then calls the score_fit tool, which only does the arithmetic. The same evidence always gives the same score, and no AI runs on our server: the judgement stays with your own assistant. The tool receives short requirement names and levels, never your profile or evidence.

VerdictWhenWhat happens
Strong80 or more, no missing blockerWrites the ATS resume.
Partial50 to 79, or a non-core blocker is missingWrites the resume with the gaps left out, and gives a plan for the gaps.
Not yetUnder 50, or a core blocker is missingNo padded resume. Gives an honest explanation, an improvement plan and fitting roles.

You can also ask it to rank several roles. It returns a table sorted by score with the biggest gap for each.

Your profile

This is the JSON the assistant builds and shows you to confirm. You can write it yourself and paste it, or keep it as resume-profile.json. Leave unknown fields empty; it will not guess.

resume-profile.json
{
  "name": "Asha Rao",
  "email": "[email protected]",
  "phone": "+91 90000 00000",
  "location": { "city": "Pune", "country": "India" },
  "links": { "linkedin": "https://linkedin.com/in/asha", "github": "https://github.com/asha", "portfolio": "" },
  "goal": "Backend engineer at an early-stage startup",
  "education": [{ "school": "Example University", "degree": "B.Tech", "field": "Computer Science", "start": "2021", "end": "2025", "grade": "8.4 CGPA" }],
  "experience": [],
  "projects": [{
    "name": "Taskly",
    "github": "https://github.com/asha/taskly",
    "live": "",
    "role": "Solo",
    "stack": ["Node.js", "Express", "PostgreSQL"],
    "what_it_does": "To-do app with accounts and reminders",
    "facts_from_repo": ["REST API with Express and PostgreSQL", "has a Dockerfile", "no tests found"]
  }],
  "certifications": [],
  "skills": { "languages": ["JavaScript", "SQL"], "tools": ["Git", "Docker"], "methods": [] }
}

It never asks for a photo, date of birth, marital status, ID numbers or a full home address. City and country are enough.

Your resume files

When the verdict is Strong or Partial, the skill saves the finished resume as real files in one folder, so you never copy text out of a chat. It creates a folder called resume-output in your working directory, containing:

  • Firstname-Lastname-Role.docx: the resume in Word format, the safest for applicant tracking systems.
  • Firstname-Lastname-Role.pdf: the same resume exported from that Word file, with selectable text.
  • resume-profile.json: your confirmed profile, so the next role takes seconds.
  • cover-letter.docx: only if you ask for one.

The layout is one column, A4 or Letter, a single standard font, real headings and bullets, and no tables, images or text boxes. If your assistant cannot create files, it gives you the resume as clean text to paste into Word.

See what you get: sample resume (.docx) and sample resume (.pdf). The person and company are made up.

Example: not ready yet

A student with a simple to-do app asks about a role that needs Kubernetes, Docker and load balancing. The skill answers like this:

Assistant
Sorry, I can't honestly build a strong resume for this role yet.
It asks for Kubernetes, Docker and load balancing, and your projects
(a to-do app) don't show any of them. I won't add skills you haven't
used, because they get tested in interviews.

You can close the gap. Here is how to make your project stronger:
  Week 1      Dockerfile + docker-compose.yml, add tests
  Weeks 2-3   Run it on Kubernetes locally (kind or minikube),
              put nginx or an ingress in front of two replicas
  Week 4      README with an architecture diagram, live link, CI

Roles that fit what you have today: ...

Once you have built it, come back. It re-reads your repo, re-runs the match and writes the resume then.

What it will not do

  • Invent experience, skills, numbers, dates, degrees or certifications.
  • List planned or in-progress learning as a skill.
  • Promise that a resume will pass an ATS or get you an interview. It improves the odds of being read correctly.
  • Give apply links. Those are a Premium feature of the site; it links to the role page instead.

Privacy

Your profile and resume stay in your own conversation with your assistant. The skill sends nothing about you to JoinAStartUp: the MCP tools receive only the job link, slug or search words you ask about. See the Privacy Policy for what the site itself collects when you sign in to the MCP server.

Troubleshooting

ProblemFix
I cannot find the career_copilot promptCheck that the MCP server is connected and that your account is Premium. Reconnect it from the MCP page and refresh the prompt list. Some assistants do not show MCP prompts; ask in plain words instead (see How to use).
It stopped working in the middle of a conversationYour Premium may have ended. Check your account, upgrade if needed, then reconnect the MCP server from your assistant.
It cannot read the jobConnect the MCP server from the MCP page and sign in with a Premium account. Paste the job text as a fallback.
It cannot read my repoThe repo may be private. Paste the README and the file list, and it will say that is what it used.
No files appearYour assistant may not be able to create files. It then gives the resume as clean text to paste into a Word document.
The verdict feels too harshIt only counts what is proven. Add the missing evidence (tests, a Dockerfile, a live link, your role in a team project) and ask again.

What the prompt contains

Choosing career_copilot loads these instructions into your conversation:

  • The overall flow and the honesty rules.
  • How your profile and projects are collected.
  • Job match, with the scoring rubric.
  • The ATS resume writer, its layout rules and the file checklist.
  • Suggestions for closing gaps.
  • Short guides on ATS facts and on writing bullets.

Some ideas in the ATS and bullet-writing guides (ATS myths and truths, adding the domain to each role, the delivery checklist) were adapted from resume-builder-skill by dabydat, MIT licensed.

Questions or something missing? Use the contact page.