// built for agents
Hand freehire
to your agent.
Millions of open postings behind a public API with no login wall, so an agent can run the whole hunt: search, shortlist, track, reframe a CV, read the replies. There is no sign-up to get through first — the fastest way to start is to paste the block on the right into whatever agent you already use.
Searching needs no key at all. One is only needed for the half that belongs to your account — saving, applying, stages, CV work and application mail.
Install freehire and use it to run my job search. curl -fsSL https://freehire.me/install.sh | sh Then read https://freehire.me/llms.txt for the conventions, and `freehire --help` for the commands.
↑ paste into your agent — it takes it from there
// pick your track
Three ways in, and they are not equal.
All three run on one fhk_ key and one account.
What separates them is how much of the surface the host can actually reach.
A local harness
Claude Code, Codex, Cursor, Cline, Aider — anything that runs commands on your machine. Give it the hand-off block and it installs the CLI itself. This is the only track that reaches the whole surface, because it is the only one that can also touch your files: reading the CV you keep locally, writing the tailored PDF back next to it.
The CLI ships five agent skills named for the task rather than the tool, so a host loads what the question needs instead of the whole surface — plus a Claude Code plugin that brings their slash commands with them.
# no Go toolchain needed curl -fsSL https://freehire.me/install.sh | sh # only for the account half freehire auth login --token fhk_… # vocabulary first, then search freehire facets freehire search "golang" --remote --region eu
Claude Desktop, or any MCP host
freehire-mcp exposes search, market fit,
tracking and CV tailoring as Model Context Protocol tools. It runs via npx — no global install —
and reuses the CLI credentials if you have already signed in there.
{
"mcpServers": {
"freehire": {
"command": "npx",
"args": ["-y", "freehire-mcp"]
}
}
} ChatGPT
A published custom GPT is already wired to the same API, and searching in it needs no setup at all. Paste your own key into its authentication field and it can save, apply and move stages too. Prefer to build your own? The OpenAPI schema imports straight into an Action.
What it cannot do is the work that needs your files: there is no CV rendering and no mail triage on this track.
Rolling your own? Every page on this site is
one unauthenticated JSON call. Start from the API reference or the OpenAPI schema ↗, and hand your agent llms.txt ↗, which states the conventions in
the form it reads best. Please send a User-Agent that names you.
// what your agent can do
Search grounded in facets
Ask for the live filter vocabulary first, then search with real values instead of guesses. A filter nothing reads is ignored, not refused — so a typo silently returns everything.
Advanced searchScore a stack against demand
Market fit measures how much of the open market your skills already cover, and puts a number on every gap — which is a better question than "am I qualified for this one job".
Market insightsTrack every application
Save, mark applied, move a stage, attach a note. Everything lands on the same board the website shows, because it is the same account.
Application trackingTailor a CV to one vacancy
Reframe what you actually did toward one posting, cite every claim, invent nothing, then render an ATS-ready PDF. The agent is told which requirements your history covers and which it does not.
CV tailoringTriage the replies
Push a batch of application mail, judge each message, and drain the two link queues. A confident verdict moves the application forward on its own.
InboxSkip the dead postings
Every posting carries a reality signal, so an agent can drop the ones that have gone stale instead of writing you a cover letter for a role nobody is filling.
Ghost jobsAn agent never submits an application for you — you still apply on the employer's own site and freehire records that you did. The one exception works the other way round: the browser extension fills an ATS form for you to check before you press send.
// why local wins
The surface is the difference
A hosted assistant runs freehire's tools inside someone else's conversation. A local harness runs them next to your files, which is where a job hunt actually lives: the CV you keep, the notes you take, the mailbox you export. That is the whole of the gap — not model quality.
Every mark below comes from the client's own tool list. If a surface gains a tool, this table changes with it.
| Capability | Local CLI | MCP host | ChatGPT |
|---|---|---|---|
| Search, facets, market fit | ● Supported | ● Supported | ● Supported |
| Save, apply, stages, notes | ● Supported | ● Supported | ◐ Supported once you paste an API key |
| CV tailoring and PDF export | ● Supported | ● Supported | – Not available on this surface |
| Application mail triage | ● Supported | – Not available on this surface | – Not available on this surface |
| Reads and writes your own files | ● Supported | – Not available on this surface | – Not available on this surface |
| Task-shaped agent skills | ● Supported | – Not available on this surface | – Not available on this surface |
● supported · ◐ needs your API key · – not on this surface
// faq
- Do I need an API key to let an agent search freehire?
- No. Job search, company data and facets are public and need no authentication, so an agent can search the whole catalogue the moment it is installed. A key is only needed for the parts that belong to your account: saving, applying, application stages, notes, CV tailoring and application mail.
- Which kind of agent works best with freehire?
- A local harness — Claude Code, Codex, Cursor, Cline and the like — driving the freehire CLI. It gets the whole surface, including CV edits and PDF rendering, it can read and write files on your machine, and nothing caps how many tool calls one task may make. Hosted assistants come next: an MCP host such as Claude Desktop runs the same tools over the Model Context Protocol.
- How do I connect freehire to Claude?
- Two ways. In Claude Code, install the CLI and its agent skills, or add the plugin marketplace. In Claude Desktop, add the freehire MCP server — it runs via npx, needs no global install, and reuses the CLI credentials if you have already signed in.
- Can ChatGPT use freehire?
- Yes. There is a published custom GPT wired to the freehire API, and searching in it needs no setup. To save, apply and track, paste your own API key into the GPT authentication field. You can also import the OpenAPI schema into an Action of your own.
- Can my agent apply to jobs for me?
- No, and that is deliberate. Applications are submitted on the employer’s own site by you; freehire records that you applied, moves the application through its stages and reads the replies. The browser extension can fill an application form for you to review, but it never submits one on its own.
- Is there a rate limit for agents?
- Yes, and every response reports where you stand in the X-RateLimit-Remaining header. Ordinary reads get 600 requests per minute and the agent search endpoint gets 300. Read the header and pace yourself rather than waiting for a 429.
Free and open source — no tracking, no lock-in, and nothing to sign up for before your agent can search. Read every line of the CLI ↗ and the MCP server ↗.