The Remoet MCP server is live (free tier)

It's open.

Plug your AI agent into Remoet and let it do the boring half of looking for work. Searching companies that match your stack. Pulling jobs from those companies. Filtering by salary, location, remote policy, whatever you care about. Saving the good ones with notes. Updating your profile, work history, and projects through conversation.

You stop clicking. You start prompting.

The free tier is live in production today. The paid tiers are still gated while we wait on a payment-processor approval, so the Pro and Max plans are not turn-on-able yet. They'll show up when they're real. Until then, Free is the whole product.

If you missed the priming email

Quick recap. Remoet is a job platform built for the agent era. Companies and jobs are the data, but the way you interact with that data is your AI agent. Instead of bouncing between tabs, you talk to the agent you already use, and it talks to Remoet on your behalf.

The bridge that makes that possible is MCP, the Model Context Protocol. It's the open standard from Anthropic that lets AI clients like Claude, Cursor, and Windsurf connect to external tools and call them on your behalf. If MCP is new to you, start here. If it's not, the only thing you need to know about Remoet's implementation is the URL and the API key.

What you can actually do today

I want to be careful here because there's a lot of MCP marketing copy floating around that overpromises. Here's the real list. These all work. They've been running on staging for weeks and the bugs that mattered are out.

Find companies by your stack. Tell your agent the technologies you care about. It searches Remoet's catalogue, returns the companies that overlap, and ranks them by job count or alphabetically or however you ask. Tech stack on Remoet is per-company, scraped and enriched, not a tag a marketer wrote. That's the part nobody else has.

Star the ones you'd actually work for. Stars on Remoet are not bookmarks. They're a noise filter. Your agent uses them to scope every future job query down to the companies that match your stack. One star is enough to start. Star a single company you'd seriously work for and the daily loop already works: the agent pulls that company's jobs, ranks them, you read what matters. Star ten and the loop just covers more ground. The point is to be selective, not to fill a number.

Pull jobs from your starred companies. Filter by salary range, remote policy, experience level, location, posting age. Sort by recency or salary. The agent paginates, summarizes, ranks. You read the ones it surfaces. This is the loop most users will run daily.

Save the good ones with notes. Same as the in-app save action, but the agent does it. "Save this one as a strong yes, note that the Go salary band is below my floor but the team looks great." Read those back later.

Manage your profile by talking. Update your summary. Add work experience. Add a project. Add education. Edit a link. Your agent reads what you tell it, normalizes the dates and structure, and writes it back. The classic "I should update my profile but I never do" problem solves itself when the cost of an update drops to one sentence.

Read your weekly digests. If you've got digests on, your agent can pull them and walk you through them.

Spin up link trees. One per free account. The new dev-flavored Linktree-style page that lives at remoet.dev/u/your-handle, sourcing from your profile data so it stays current.

There are more tools than that. I'm not going to count them in a blog post because the number drifts every week. The above is the daily-driver shape.

What does not work yet

I'd rather you find this out from me than from a tool error.

You cannot apply to most jobs through the agent. Almost every job in Remoet right now is scraped from a company's external careers page. The agent can find the job and hand you the link, but the actual application happens on the company's site like always. Internal applications (where you apply through Remoet end to end) only work for companies posting directly through our partner system, and that's a tiny slice of the catalogue today. Growing, but tiny.

There's no cover-letter writing tool. Not yet. The plumbing exists, the front-end's not finished, so the MCP path skips it.

There's no GitHub integration. Your tech stack is what's in your Remoet profile, not what's actually in your repos. If you've never filled in your stack, the agent has nothing to match against. Take five minutes, fill it in, and the rest gets a lot smarter.

Filters only apply to jobs from starred companies. You can't currently ask the agent to filter the entire Remoet catalogue by salary or remote policy. The filtering happens at the "jobs from companies you've starred" layer, which is the layer where the data is rich enough for filters to be meaningful. If you want a wider net, star more companies.

That's the honest version. Everything else in the platform UI works at a similar level through the agent, with these gaps.

How to connect

Five minutes if you've used MCP before. Ten if you haven't.

1. Generate an API key

Sign in to Remoet, go to your settings, generate an MCP API key. Copy it. It looks like rmt_.... Don't paste it anywhere shared.

2. Drop the config into your agent

Claude Code. Add the server with the CLI:

claude mcp add remoet \
  --transport http \
  --url https://api.remoet.dev/mcp \
  --header "X-API-KEY: rmt_your_key_here"

Restart Claude Code after adding (it picks up new servers on launch).

Cursor. In ~/.cursor/mcp.json (or your project's .cursor/mcp.json for project-scoped):

{
  "mcpServers": {
    "remoet": {
      "type": "http",
      "url": "https://api.remoet.dev/mcp",
      "headers": { "X-API-KEY": "rmt_your_key_here" }
    }
  }
}

Reload Cursor.

Windsurf. Same JSON shape, in Windsurf's MCP settings.

Claude Web (custom connector). No API key, no JSON. Add a custom connector pointing at:

https://api.remoet.dev/mcp/oauth

Claude Web walks you through OAuth. You sign in to Remoet, approve the scopes, and you're connected. This is the lowest-friction path if you don't want to mess with config files. Custom connectors are free-tier-eligible on claude.ai, so a free Claude account is enough.

Other clients. Anything that speaks Streamable HTTP MCP can connect using the same shape as Cursor's config. If your client only supports stdio, you're out of luck for now.

3. Ask the agent something

A short script you can paste straight in:

Search Remoet for companies that use TypeScript and Postgres. Show me the top ten by job count. Star the three I'd most likely want to work at given my profile. Then pull the jobs from those three companies, filtered to remote and senior level, sorted by recency.

That single prompt exercises five tools. If everything's wired up, you'll get a shortlist in under a minute.

What the free tier actually gives you

I'm going to put the limits in plain numbers because vague tier copy helps nobody self-select.

• One star is enough. Star a single company and the agent loop is already running. No five-star activation gate, no waiting room. The active-star cap on free is ten, which gives you room to grow a real shortlist without pretending it's a bookmarking tool.

• 5 unstars per month. Stars are a commitment. Unstarring costs budget so people don't churn through the catalogue. If you hit the unstar wall, that's the signal that you're starring too aggressively.

• 30 MCP requests per day. A "request" is one tool call. The example prompt above costs about five. Most daily-driver sessions land between five and twenty.

• 300 API requests per day. Separate quota for direct API users (calling the GraphQL API yourself, not going through your agent). Most MCP users won't touch this.

• 1 link tree. Your dev-flavored profile page.

• Job data delayed by one week. This is the only "waiting room" gate. Real-time job data is a paid-tier feature. On free, the jobs your agent sees are last week's snapshot. Plenty for picking targets, slow for first-mover.

If those limits don't fit your usage and you want a louder sense of when paid lands, hit the Discord. I'd rather hear it from real users than guess at thresholds.

Why soft launch

Two reasons.

The payment piece is not approved yet. Pro and Max are built. The Paddle merchant approval is in flight, and until that lands I cannot actually charge anyone. Rather than wait on an external dependency to ship the entire MCP server, I split the launch. Free works today. Paid lights up the week we get approval.

I want to find the rough edges with real agents before opening the floodgates. A staging instance, no matter how thoroughly it's tested, doesn't surface the weird stuff. The 4am tool call that fails on a session edge case. The prompt phrasing the model picks that exposes a tool description ambiguity. The free-tier limit that turns out to be too tight or too loose for the actual workflow.

So: free tier today, paid when Paddle approves, and somewhere in between I expect to ship a handful of corrections based on what you tell me.

What I want from you

If you connect, the most useful thing you can do is tell me what broke or what felt clunky. Reply to the email, ping me on Discord, open a GitHub issue on the docs repo. I read all of it personally. The agent-era job platform is not a thing you build in a vacuum. It's a thing you build in a feedback loop with the people using it.

If it works smoothly, I want to hear that too. Send me your prompt, the time it took, the shortlist it gave you. Those are the stories that go into the next round of marketing, and they always read better than anything I'd write.

The full /mcp page is at remoet.dev/mcp with example prompts, the install snippets, and the FAQ. The agent job-search workflow guide is the deep walk-through of the daily loop.

That's the launch.

Plug in. Tell me what breaks.

Happy hunting.