Job search is a good fit for AI assistants: you describe what you want in plain words, the assistant asks the right follow-up, and you get a short list instead of 40 tabs. The weak spot is data. A model on its own will happily invent a vacancy, a salary or a company.

Jobiba is a job site that also runs a public Model Context Protocol (MCP) server, so an assistant can read live listings and real pay data instead of guessing. This post shows how to connect it in a couple of minutes, and what it does and won't do.

Two servers, one for each side of hiring

URL For
https://jobiba.com/mcp/jobs Job seekers: search, salaries, CV, apply, alerts
https://jobiba.com/mcp/hiring Employers and recruiters: post jobs, screen applicants, shortlist
https://jobiba.com/mcp Everything in one

They are split on purpose. Each address is its own OAuth resource with its own tools, so a candidate's assistant never sees employer tools, and a token issued for one address is refused by the other.

Job search, job details, salary research and company research work without an account. Tools that read or change your own data (your CV, applications, alerts) ask you to sign in once.

Connect it

ChatGPT: Jobiba is in the ChatGPT app directory. Open it, press Connect, sign in when asked, then pick Jobiba from the + menu in a chat.

Claude (web, desktop, mobile): Settings → Connectors → Add custom connector → https://jobiba.com/mcp/jobs.

Claude Code:

claude mcp add --transport http jobiba https://jobiba.com/mcp

Cursor, VS Code, Windsurf (mcp.json):

{ "mcpServers": { "jobiba": { "type": "http", "url": "https://jobiba.com/mcp/jobs" } } }

Gemini CLI:

gemini mcp add --transport http jobiba https://jobiba.com/mcp

Things to try

  • "Find remote data analyst jobs that list pay in USD."
  • "What does a registered nurse earn in London? Show the sample size."
  • "Here is my offer: ₹14 LPA for a backend developer in Pune. Is it fair?"
  • "Tailor my CV to this job and show me what changed before I apply."

The salary answers come from pay stated in live job ads, with the sample size, and say so. If there isn't enough data, the tool says that instead of filling the gap.

Guardrails worth copying

If you are building your own MCP server, these rules saved us from the usual failure modes:

  1. Facts only from the data. The assistant never invents jobs, salaries, employers or your experience.
  2. Preview, then confirm. Applying, messaging a candidate or moving an applicant to another stage returns a preview first and acts only after the user says yes.
  3. CV tailoring can't add facts. It rewrites and reorders what is true; it never adds an employer, title, date, degree or skill you didn't confirm.
  4. No payments in chat. Upgrades open a billing page on the site.
  5. Contact details stay private. Employers don't see a candidate's email or CV file in chat unless the candidate applied or agreed.

No assistant? There's a plain REST API too 

curl 'https://jobiba.com/api/v1/jobs?q=nurse&location=London&remote=HYBRID&limit=20'

No key, HTTPS, CORS on. Filters: q, location, type, remote, since, page, limit (up to 100). The contract is at jobiba.com/openapi.json.

Links

Bhagwati Suthar is the founder of Jobiba (Jobiba AI Labs).