MCP/CLI
Your favourite agents can already take this over.
Your whole job search, done inside the chat you already have open.
setup message · paste into your agent
Copy this for your agent
Paste it into the chat. It reads the right docs, wires itself up, and reports back.
Set up the AI Applyd MCP server in the client you are running inside, then prove it works.
WHAT IT IS
AI Applyd is a job-search agent. Its MCP server gives you tools to rewrite a resume against one specific posting, pull the real requirements out of a job description, draft a cover letter, build interview prep, translate a resume, search jobs matched to my profile, and start an auto-apply that files a real application on the employer's own hiring system.
WHERE IT LIVES
Server name: aiapplyd
Transport: Streamable HTTP
URL: https://mcp.aiapplyd.com/mcp
There is no API key, no bearer token and no client secret. Do not ask me for one, because none exists. Do not wrap this in npx mcp-remote. Do not use an SSE endpoint.
HOW TO REGISTER IT
Use the line for the client you are actually running inside, at user scope, so it is still there in my other projects.
Claude Code:
claude mcp add --transport http --scope user aiapplyd https://mcp.aiapplyd.com/mcp
Codex CLI:
codex mcp add aiapplyd --url https://mcp.aiapplyd.com/mcp
Cursor, in ~/.cursor/mcp.json:
{ "mcpServers": { "aiapplyd": { "url": "https://mcp.aiapplyd.com/mcp" } } }
Cline, in ~/.cline/data/settings/cline_mcp_settings.json:
{ "mcpServers": { "aiapplyd": { "type": "streamableHttp", "url": "https://mcp.aiapplyd.com/mcp" } } }
The type field is not optional there. Leave it out and Cline falls back to the deprecated transport without telling anyone.
Any other client: read that client's own current MCP documentation and register a remote Streamable HTTP server the way it expects today. Do not rebuild a config shape from memory. These files moved several times this year.
HOW THE SIGN-IN WORKS
OAuth 2.1 with PKCE and dynamic client registration. Nothing runs anonymously and most clients do nothing at all at add time. The first tool call comes back 401 on purpose, a browser opens at mcp.aiapplyd.com, I sign in with Google or a magic link, I approve a consent screen that names the client asking, and your client stores the token. Claude Code can get it over with up front: claude mcp login aiapplyd
HOW TO VERIFY, IN THIS ORDER
1. List my MCP servers and confirm aiapplyd is registered and connected.
2. List its tools and confirm you can see names beginning with aiapplyd_.
3. Call aiapplyd_search_jobs once with a short query.
4. If that comes back 401, I have not signed in yet. Tell me to finish it in the browser, then call the tool again.
5. Tell me which step you reached and what came back.
An empty tool list before I sign in is expected. It is not a failure and it is not a reason to start editing other files.
RULES
Do not touch, reorder or remove any other MCP server in my configuration.
Do not add an Authorization header. Some clients report the whole connection as failed instead of falling back to OAuth.
Do not invent a config key you have not read in that client's documentation.Claude CodeClaudeChatGPT/CodexCursor
Cline too, and anything else that speaks MCP. Free account, no card, and you sign in the first time it acts for you.
Setting it up by hand? The address is
https://mcp.aiapplyd.com/mcpEverything you can ask for, in the chat.
Your free account covers matching, resumes and interview prep. Applying and cover letters are on the paid plans.
Paid plan
It applies for you
Point it at a job. It fills the form on Greenhouse, Workday or Lever, in your words, and presses send.
Nothing counts as sent until the employer confirms it back.
Free account
Your resume, rewritten for one job
Rewritten against the posting you name, from work you have really done. Any language.
Free account
Jobs worth your evening
Ranked against what you have actually done, not against your job title.
Free account
Ready before the interview
The questions this role will ask, each matched to a story you already have.
Paid plan
A cover letter worth sending
Your real history, the actual company, and a filter that refuses to write "passionate about".
One agent. Two ways in.
Remote MCP works from the chat. WebMCP works when a compatible browser agent is already on AI Applyd.
Remote MCP
Connect once from ChatGPT, Claude, Cursor, or another MCP client. Your agent can use AI Applyd without keeping this page open.
Browser-native WebMCP
In Chrome’s experimental WebMCP trial, an agent visiting AI Applyd can read public jobs and plans. After you sign in, it can read your matches, job details, and application progress, then save a job for later.
WebMCP uses the browser session already open on this page. It does not replace the remote MCP server, and it does not make off-site agents discover AI Applyd before they visit.
What people ask before they start.
The AI Applyd MCP server puts AI Applyd's job-search tools inside your AI assistant, so the chat you already have open can do the work instead of describing it. MCP is the open standard for connecting AI clients to external tools. Once it is connected you can pull apart a job description, rewrite your resume against it, draft the cover letter, translate the resume, prep for the interview and search live matched jobs, all without leaving ChatGPT or Claude. The connection URL is https://mcp.aiapplyd.com/mcp, and every tool runs against your own AI Applyd account.
Any MCP-compatible client can connect to the AI Applyd MCP server. That includes ChatGPT, Claude (Desktop, Code and web), Cursor, Windsurf and Cline. Both transports are supported: Streamable HTTP at https://mcp.aiapplyd.com/mcp for modern clients, and Server-Sent Events at https://mcp.aiapplyd.com/sse for the ones still on SSE.
They are two different ways for an agent to use AI Applyd. The remote MCP server connects AI Applyd to ChatGPT, Claude, Cursor and other MCP clients even when aiapplyd.com is not open. WebMCP is an experimental browser feature: after a compatible agent visits AI Applyd, the page can offer tools through the browser session already open there. Signed-out WebMCP tools can read public jobs and plans. Signed-in tools can read your matches, applications and job details, and save a job. WebMCP is page-bound, so it does not replace remote MCP and does not create off-site discovery by itself.
The AI Applyd MCP server is free to connect, on a free account. There is nothing to buy and no card to enter: you sign in with Google once, and the account starts with 10,000 tokens. Job-description analysis, resume optimization, translation, interview prep and the job match-finder all run on that balance. Cover letters and auto-apply are on a paid plan. Free here means free account, not no account: there is no anonymous tier, and a tool call without a session gets a 401 rather than a result. Top up or upgrade in your dashboard when you want more runs.
Yes, and it is free to create. The first time you ask your assistant for a tool it opens a Google sign-in, no card. That account is what the tools read your profile from and what they spend tokens against, and it starts with 10,000 of them. Every tool is behind it, without exception: an unauthenticated call is answered with a 401 so your client knows to start the login rather than quietly returning nothing. Cover letters and auto-apply additionally require a paid plan.
On a free account, drawing on your 10,000 starter tokens: job-description analysis, resume optimization against a specific posting, resume translation, PDF export, interview question generation, the job match-finder, and your job preferences. On a paid plan: cover letter generation and auto-apply, which files the finished application on the employer's own hiring system. Each tool returns structured results your assistant can read back and act on, so "rewrite this for the Stripe posting and then apply" is one message rather than an afternoon.
Tools spend tokens from the same balance as the web app, so the 10,000 you start with cover your first runs. Tokens are spent on the real compute each run uses, so you pay for what you actually ran. Top up or upgrade in your dashboard when you get low. There is no separate MCP meter and no second bill: one balance, whether the request came from the assistant or the dashboard.
Yes. Data is encrypted in transit (TLS) and at rest, never sold, and never used to train AI on your resume. The account connection uses scoped access, so the assistant only reads what the tool it is running actually needs, and you can disconnect the client or delete every byte from your dashboard whenever you like.