Guide to Connecting an AI Job Search Agent to Claude or ChatGPT in 2026

Published on September 7, 2026 by Tsenta

If you have asked Claude or ChatGPT to help with your job search, you already know where it stops. The assistant will happily rewrite a résumé bullet or draft a cover letter, but it cannot open an employer's application portal and send anything. Closing that gap takes a purpose-built agent wired into your AI client through the Model Context Protocol. Tsenta publishes a remote MCP server at https://api.autojobs.me/api/v1/mcp that does this: it surfaces relevant roles, prepares job-specific materials from your saved candidate context, and submits complete applications on supported employer systems, with an inspectable receipt for every successful submission. This guide covers what that connection is, how to set it up in each client, and where it genuinely falls short.

What an AI Job Search Agent Connected to Your Assistant Actually Does

An AI job search agent is the layer that does the mechanical work of applying: finding matching roles, preparing application materials for a specific posting, filling in the employer's form, and submitting it. The operative word is submitting. Most tools marketed in this category generate text or pre-fill a handful of fields and then leave the last mile to you, which is the part that actually takes the time.

Connecting that agent to Claude or ChatGPT means exposing it through MCP, the open standard Anthropic introduced for letting AI clients call external tools and services from inside a conversation. Think of it as a common plug. Once a job-application agent publishes an MCP server, any compatible client can call it in natural language. You type something like "find senior product manager roles at fintech companies and apply to the strongest matches," and the connected agent carries that instruction through to a real submission rather than writing you a plan for doing it yourself.

Why an MCP Server for Job Applications Beats a Chatbot That Gives Advice

The value is not that your assistant learns about jobs. It is that the assistant gains the ability to act, and that the acting happens where you are already typing.

Timing is part of why this matters. Applications cluster hard toward the start of a posting's life, and a 2017 St. Louis Fed analysis of roughly 80 million applications found that 41 percent of applications arrive within the first 48 hours of a posting and 56 percent within the first 96 hours. If applying means setting aside an evening to work through portals by hand, you are structurally late to a lot of roles. An agent you can invoke in a sentence removes the scheduling problem, because the cost of applying drops to the cost of asking.

The hiring side is moving the same direction. Korn Ferry's research on talent acquisition trends for 2026 found that 52 percent of leaders plan to add autonomous agents to their teams in 2026. Candidates working entirely by hand against increasingly automated screening are giving up ground for no particular reason.

What makes MCP the right mechanism, rather than a bespoke integration per client, is that it is a standard. You configure one endpoint and it works across clients, which means switching from Claude to Cursor does not mean relearning your job search setup.

What You Need Before You Connect

Setup is quick, but the agent is only as good as what you give it. Get these in order first.

Before you paste any config:

• A Tsenta account with your candidate context filled in, including work history, education, work authorization, and role preferences
• A current résumé uploaded, since job-specific materials are prepared from it
• An AI client that supports MCP, such as Claude Code, Claude Desktop, Claude on the web, Cursor, or Codex CLI
• On claude.ai, any plan tier works, since custom connectors are available on Free, Pro, Max, Team, and Enterprise, though a free account can hold only one custom connector at a time

The résumé point is worth dwelling on. Everything the agent prepares is grounded in the context you saved, so vague project descriptions and a half-finished work history produce weaker applications no matter how good the instruction you type. Thirty minutes spent on the source document pays for itself across every application afterward.

AI Job Search Tools That Work with Claude

Claude has two setup paths depending on where you use it, and both take about a minute.

Claude Code

Claude Code is Anthropic's terminal agent, and adding Tsenta is a single command:

claude mcp add --transport http tsenta https://api.autojobs.me/api/v1/mcp

Run it, then approve the OAuth screen that opens in your browser. Confirm it registered with claude mcp list, which should show tsenta as connected. Claude Code stores a short-lived OAuth token locally and refreshes it automatically, so you never handle the token yourself.

Claude Desktop and Claude on the Web

Claude Desktop and claude.ai share the same Connectors panel, so the steps are identical in both, with no config files and no CLI. Open Settings, go to Connectors, choose to add a custom connector, and paste https://api.autojobs.me/api/v1/mcp as the URL. Approve the OAuth prompt, then look for the Connected badge next to Tsenta in the list. Start a new chat and the Tsenta tools will be available to it.

Claude manages the token silently from there. If you sign out of Tsenta or revoke access, Claude will prompt you to reconnect rather than failing quietly.

How to Connect ChatGPT to an Auto-Apply Agent

Because MCP is an open standard rather than a vendor-specific plugin format, the same endpoint works anywhere the protocol is implemented. In ChatGPT, that means adding Tsenta as a custom MCP connector using https://api.autojobs.me/api/v1/mcp and approving the OAuth flow, on a plan and mode where custom connectors are exposed.

Worth being straight about: Tsenta publishes tested, step-by-step setup instructions for Claude Code, Claude Desktop and web, Cursor, and Codex CLI. ChatGPT's connector surface has moved more than once, so follow your client's current documentation for where custom connectors live and use the Tsenta URL there rather than copying a stale screenshot from anywhere. The endpoint and the OAuth flow are the same in every case.

A Job Application Agent for Developers Using AI Assistants

If you already work in an editor or a terminal, the connection is config rather than clicks.

Cursor reads MCP servers from a JSON config file. Open Cursor's MCP settings and add Tsenta as a server entry pointing at https://api.autojobs.me/api/v1/mcp, then restart Cursor and approve the OAuth prompt. Cursor keeps the token in its secure credential store, and the config file itself holds only the URL, never a secret. To confirm it worked, open a chat and ask what Tsenta tools are available.

Codex CLI reads MCP servers from a TOML config at ~/.codex/config.toml. Open that file, append a server block naming Tsenta with the same URL, then start Codex and approve OAuth. You should see tsenta listed as connected along with its tools.

The developer case is not just convenience. Because MCP exposes the agent's tools in a standard, machine-readable way, you can compose Tsenta into a larger workflow you build yourself, calling specific capabilities from your own multi-step setup instead of driving everything through a chat window.

Client	Setup method	Approx. time	Notes
Claude Code	One CLI command	Under a minute	Verify with claude mcp list
Claude Desktop	Settings, then Connectors	Under a minute	No config file needed
Claude on the web	Settings, then Connectors	Under a minute	All plans; Free is capped at one connector
Cursor	JSON config entry	A few minutes	Restart required after editing
Codex CLI	TOML config block	A few minutes	Config lives at ~/.codex/config.toml
ChatGPT	Custom MCP connector	Varies	Follow your client's current connector flow

What to Look for in an Agent Before You Wire It In

Plenty of tools now advertise MCP support, and the phrase covers a wide range of usefulness. These are the questions that actually separate them.

Does it submit, or only search?

If the agent returns a list of matching roles into your chat and stops there, you have added a search box to your assistant, not an application agent. Check for end-to-end submission on real employer systems specifically.

How many employer systems does it cover?

Coverage is the ceiling on everything else. Tsenta submits on 30+ supported employer systems, including Workday, Greenhouse, and Lever, which matters because candidates apply across many different portals rather than one. A tool that only works on a single job board is solving a narrower problem than the one you have.

Are materials prepared per role?

Volume without relevance is just noise. Materials should be generated from the actual posting and your real background, not sent as one generic file everywhere.

Can you see what was sent?

This is the trust question. Tsenta makes review controls available and produces an inspectable receipt for every successful submission, so you can check what went out under your name. An agent that submits invisibly puts your professional reputation somewhere you cannot inspect.

Is the pricing published and capped?

"Unlimited" usually means a limit that is written down somewhere you have not read. Defined caps with published prices are a better signal.

How People Actually Use This

The candidates who get the most out of an assistant-connected agent tend to be running high-volume searches where applying by hand cannot keep pace.

New grads and career switchers applying across a hundred or more roles in a cycle use it to compress the repetitive part. From a single chat, they ask for relevant roles in their target function, review what comes back, and approve applications without opening a separate portal for each one.

International students on F-1 or OPT have a narrow authorization window and need to cover a lot of ground inside it. Tsenta surfaces the sponsorship signals it knows about and answers work-authorization questions consistently across employer forms. It does not guarantee any company's sponsorship status, which is a limit worth stating plainly rather than a detail to bury.

Software engineers already working in Claude Code or Cursor tend to use it without changing context at all, running a search in the same session as their actual work and checking receipts inline.

Candidates in data, product, finance, and consulting apply across several employer systems at once, since roles in those functions get posted on different portals depending on the company. Driving all of them through one instruction is materially less tedious than keeping five tabs open.

Getting Better Results Once It Is Connected

Connecting the agent is the easy part. These habits are what separate useful volume from noise.

Be specific in your instructions. "Find remote data engineering roles at Series B or later startups in the US and apply to the strongest matches" gives the agent far more to work with than "find me data jobs." Role, seniority, location, and company stage all narrow the set to roles worth your attention.

Keep your saved context current. When you finish a project or change titles, update it, because every application afterward is built from that record.

Use review controls where the role matters most. For a job you genuinely care about, spending a minute checking before it goes out is worth it. For breadth applications, lighter oversight is a reasonable tradeoff you get to make deliberately.

Check your receipts. They exist so you have a record of what was submitted, which is useful both for follow-up and for noticing when something about your saved context needs fixing.

Match your tier to your real volume rather than your optimistic one, since allowances do not roll over.

Honest Limitations

Anything this article does not tell you, you will find out in week two, so here it is up front.

Coverage is finite. Tsenta submits on 30+ supported employer systems. That covers a lot of where jobs live, but a role posted on a custom in-house portal outside that set will not be submitted automatically, and no amount of instruction to your assistant changes that.

Free plans on claude.ai get one connector slot. Custom connectors are available on every tier, including Free, but a free account can hold only one custom connector at a time, so Tsenta has to be the one you pick. On Team and Enterprise plans, an Owner has to add the connector to the organization before members can enable it individually. Plan rules verified September 2026 against Anthropic's connector documentation, which has changed before, so check it if something does not match.

Output quality tracks input quality. The agent prepares materials from your saved context. A thin profile produces thin applications, and the agent will not invent experience to compensate, which is correct behavior but does mean the setup work is not optional.

Completion time varies by employer system. Different portals have different flows, some with multi-step forms and open-ended questions. Some applications finish quickly and others take longer, so treat any promise of uniform instant submission with suspicion.

Employers still decide. An agent changes how efficiently you apply. It does not change screening criteria, sponsorship policy, or whether a company wants to hire you. Anyone selling the second thing is selling something else.

ChatGPT setup is less documented than Claude setup. Tsenta's tested instructions cover Claude, Cursor, and Codex CLI. ChatGPT works through the same standard endpoint, but you are following your client's connector docs rather than a Tsenta-authored walkthrough.

Pricing

Pricing is published and capped, with no add-on required for the features that matter.

The tiers:

• First 25 applications free, no card required
• Starter: $19 for 600 applications per 30-day cycle
• Pro: $39 for 1,500 applications per 30-day cycle
• Power: $99 for 4,500 applications per 30-day cycle

Two details do more work than the headline numbers. Only successful submissions consume your allowance, and skipped or failed applications are refunded, so you are not paying for attempts that did not land. Monthly allowances do not roll over, which is the reason to pick a tier that matches how you actually search rather than the busiest week you can imagine.

Where This Is Heading

The interesting shift in 2026 is not that AI assistants can discuss your job search. They could do that two years ago. It is that the plumbing for letting them take real action became standard, so the question moved from whether you can connect an agent to which agent is worth connecting.

That is the distinction to apply when you evaluate anything in this category. An assistant wired to a search layer still leaves you doing the applying. An assistant wired to an agent that submits on supported employer systems, prepares materials per role, and hands back a receipt has actually taken work off your plate. Tsenta, part of Y Combinator's S26 batch, is built for the second case, and it is reachable from the MCP server, the web dashboard, iOS, and Android, so the same agent follows you across the places you already work. If Workday portals are the specific thing slowing you down, the breakdown of AI agents for Workday applications goes deeper on that.

Setup takes about a minute, and the first 25 applications are free with no card. Connect it, give it one real instruction, and see whether the receipts look like something you would have sent yourself.

FAQs About Connecting an AI Job Search Agent to Claude or ChatGPT

What is an AI job search agent that connects to Claude or ChatGPT?

It is a job-application tool that plugs into your AI assistant through the Model Context Protocol, an open standard for letting AI clients call external tools. Once connected, you can ask your assistant to find roles and apply in plain language, and it calls the agent instead of just writing advice back to you. Tsenta publishes a remote MCP server for exactly this, so the assistant can surface relevant roles, prepare job-specific materials from your saved context, and submit complete applications on supported employer systems. Every successful submission produces an inspectable receipt, so you can check what went out in your name.

How do I connect Tsenta to Claude?

In Claude Code it is one command, claude mcp add --transport http tsenta https://api.autojobs.me/api/v1/mcp, followed by approving the OAuth screen in your browser. In Claude Desktop and Claude on the web, open Settings, go to Connectors, add a custom connector, and paste the same URL. You then approve access and look for the Connected badge next to Tsenta before starting a new chat. Claude handles the token refresh silently, and the config you paste holds only the URL, never a secret.

Can I connect ChatGPT to an auto-apply agent?

Yes, where your ChatGPT plan exposes custom MCP connectors. Because MCP is an open standard rather than a vendor-specific plugin format, the same Tsenta endpoint that works in Claude works in any client that implements the protocol. You add it as a custom connector, paste the URL, and approve the OAuth prompt. Tsenta publishes step-by-step setup for Claude Code, Claude Desktop and web, Cursor, and Codex CLI, so ChatGPT users should follow their own client's current connector flow and use the same endpoint.

Does the agent actually submit applications, or just find jobs?

This is the distinction worth checking before you wire anything in, because plenty of tools stop at search. Tsenta submits complete applications on 30+ supported employer systems, including Workday, Greenhouse, Lever, and Ashby, rather than handing you a list of links to finish by hand. It prepares job-specific materials from your saved candidate context first, and review controls are available so you decide how much oversight to apply. Completion time varies by employer system, and every successful submission produces a receipt.

Is a job application agent useful for developers using AI assistants?

It fits well if you already live in a terminal or editor, because you can drive the whole search from the same session you work in. Tsenta's MCP server connects to Claude Code, Cursor, and Codex CLI over HTTP, so a job search becomes another tool call rather than a browser tab. Developers can also compose it into their own multi-step agent workflows, since MCP exposes the tools in a standard way. The practical benefit is that applying stops competing with focused work for your attention.

What does Tsenta cost, and how is usage counted?

Tsenta gives you the first 25 applications free, and no card is required. After that, Starter is $19 for 600 applications per 30-day cycle, Pro is $39 for 1,500, and Power is $99 for 4,500. Only successful submissions consume your allowance, and skipped or failed applications are refunded, so you are not billed for attempts that did not land. Monthly allowances do not roll over, so it makes sense to match your tier to the volume you actually expect to apply at.

What are the real limitations of running a job search through an AI client?

Coverage is the main one, since Tsenta submits on 30+ supported employer systems and a role posted somewhere outside that set will not be submitted automatically. On claude.ai, custom connectors are available on the Free, Pro, Max, Team, and Enterprise plans, but free accounts are limited to one custom connector at a time, and on Team and Enterprise an Owner has to add the connector before members can enable it. Output quality depends heavily on the candidate context you saved, because materials are prepared from what you gave it. Employers still control their own forms, screening, and hiring decisions, and completion time varies by employer system.