Using the API through an AI assistant

You do not have to write code or read the API reference to work with this data. Any AI assistant that can make HTTP requests can register itself, get a key in your name and answer questions about funds, startups, rounds and people from the live database. This page is for you, the person; the assistant gets its own instructions and will read them itself.

Which assistants this works with

The one requirement is that the assistant can make HTTP requests on its own — open a URL it composed, read the JSON that comes back, and do it again. That describes every coding and agentic tool in current use:

• Claude Code, Claude Cowork and Claude Desktop with a fetch tool;
• OpenAI Codex, and ChatGPT when it has a connector or a custom GPT with Actions;
• editors with an agent mode — Cursor, Windsurf, Copilot agent mode;
• automation platforms with an HTTP step — n8n, Make, Zapier;
• anything you wrote yourself on top of an LLM with a tool that can call a URL.

Where it does not work as-is: the plain chat window of ChatGPT or claude.ai with nothing but web browsing. That browsing tool is a reader, not an HTTP client — it opens links it was given, from a search result or from you, and refuses ones the model assembled itself. A registration URL with your address in it is exactly such a URL, and so is every data query afterwards. Your assistant will tell you something like "my browsing tool will not open this URL". Skip to Chat-only assistants for what still works there.

What happens after you paste the message

1. It asks for your e-mail address

Give it the address you want the account under. The key is issued in your name, the monthly allowance is yours, and you will manage both from https://parsers.vc/account/ with this address. A personal address is fine; disposable mailboxes are refused.

2. You get an e-mail

It arrives within a minute and carries two things: a six-digit code and an Open my account button. You need only one of them.

Read the code to the assistant. That is the fast path: it exchanges the code for a key and carries on. The code is valid for ten minutes and five attempts, so if you took a while, ask it to start again rather than guessing.

Or click the button instead. Whoever holds the code can create a key in your name, and you may not want to hand that to a tool whose logs you do not control. Clicking the button signs you in at parsers.vc/account/; if the account has no keys yet, one is created and shown there ready to copy. Paste that pk_live_… string to the assistant and you are at the same place. This path also needs nothing from the assistant at all, which is why it is the one to use with a chat-only tool.

3. It has a key — and so do you

The key gives 500 credits a month, refilled on the first, and expires after 30 days because it was minted inside a conversation. When it does, the same message registers the same address again and lands on the same account. A key that does not expire, a larger allowance and the usage history are all in your account.

Asking questions

From here on you talk to the assistant the way you would to an analyst who has the database open. It knows the shape of the data from the instructions it read: what a stage is, what a signal is, that a round is usually recorded before its investors are.

Two habits will save you most of your allowance, and both are worth saying to the assistant out loud once:

• Ask for size before contents. "How many are there?" costs one credit; fetching them costs one per company. The instructions teach the assistant to check the facet counts first, but a reminder does not hurt.
• Name a budget. "Keep this under 50 credits" is a constraint it can plan around; with no budget it will page through whatever matches.

A credit buys one record, not one request: a startup or a fund is 1, a funding round is 0.5, a news item or a signal is 0.25, a search is 1 flat however many results it returns. The full card is at /api_page/pricing/. Every response tells the assistant what the call cost and what is left, so "how many credits do I have?" is answered from the last reply with no extra call.

Chat-only assistants

If all you have is ChatGPT or claude.ai with web browsing, you can still get a key, and you can still ask questions — you just do the fetching yourself.

Getting the key is the button path above. Open https://parsers.vc/api/v1/auth/register?email=your address in your own browser, click Open my account in the e-mail, copy the key the page shows, paste it into the chat. The assistant never had to call anything.

Querying works because the browsing tool will open a URL that came from you. Ask the assistant to write the URL for the question you have — it knows the syntax — then paste that URL back to it as your own message, and it will fetch and read it. The key travels in the URL as &token=pk_live_…. It is a workable way to get an answer or two; it is not a way to work all day. If you find yourself doing it more than a few times, any tool from the list above removes the step.

When the assistant gets stuck

What to know about the key

• It is yours. Revoke it at any time at /account/ — it stops working immediately. Do that if you stop using the assistant, or if the conversation it lives in is shared.
• The assistant is told not to print it into transcripts, logs or files. If you see it do so anyway, revoke and re-register; it takes a minute.
• One account per person. The assistant is told not to register a second address to extend the allowance. If 500 credits is not enough, the plans exist for that.
• Register your own address, not somebody else's. The e-mail goes there, and so does the account.

Human questions: [email protected]. The instructions the assistant reads: agents.md. The reference for people who would rather write the calls themselves: /api_page/reference/.