Zintaby Zinwave
Start free
Works with Claude, ChatGPT and Cursor

Turn the AI you already pay for into a prospecting agent

Zinta plugs straight into Claude, ChatGPT, Cursor or Codex, so finding customers becomes something you ask for in the chat you already have open. It brings the targeting, the memory of everyone you have already seen, and a bar that refuses any company it cannot show you a source for.

Connect itSee plans
$19
a month
100
companies
4
tools
0
emails sent for you

Endpoint

https://getzinta.com/api/mcp
Sign in
With your Zinta account, or a key
Setup
About two minutes
Installs
Nothing, on any machine

Nothing to install. Any MCP client that can call a URL is supported, rather than one.

Connect

Two minutes, in the tool you already use

Pick the app you use. If it is Claude or ChatGPT there is nothing to install and no key to copy: you paste an address, sign in, and approve it once.

No key

Clients with a connector UI

Paste the URL. It discovers the authorization server, you sign in and approve once, and there is nothing to copy or lose.

Bearer key

Clients with a config file

One key per account, created on the Connect screen in your dashboard. Issuing a new one replaces the old, so it is also how you revoke.

Claude's app

Desktop and web

Sign in, no key

Settings, then Connectors, then Add custom connector. You will be asked to sign in and approve once. Step by step.

Paste this as the URL

https://getzinta.com/api/mcp
Tools

Four tools, and what each is for

zinta_projects

Lists the projects this account can hunt for.

Called first when the agent does not already know which pipeline it is working on.

zinta_targeting

What this account sells, who buys it, the signals worth hunting, excluded companies, and the reasons earlier prospects were rejected.

The brief. Rejection reasons come back as rules, so the search narrows instead of repeating itself.

zinta_pipeline

The companies already in the pipeline.

Read before searching, so a hunt never surfaces the same company twice, across every session and every client.

zinta_file_prospects

Files what was found. Each row is checked before it is accepted.

A source URL that was actually fetched and whose page states the signal, a dated signal, a published address rather than a guessed one. Rows that fail come back with the reason.

Alongside

One more server in a row you have already assembled

MCP clients hold as many servers as you give them, so this is not an integration. It is a second line in the same config. If you already build through Lovable's MCP server, add Zinta beside it and both sets of tools are available in one session.

# the tool you build with
claude mcp add --transport http lovable https://mcp.lovable.dev

# and the one that finds its customers
claude mcp add --transport http zinta https://getzinta.com/api/mcp \
  --header "Authorization: Bearer YOUR_KEY"

Then the whole loop is one conversation: ship this to Lovable, then find me ten companies who would pay for it. The model reads Zinta's targeting, checks what is already in the pipeline so it never surfaces the same company twice, and files what it finds. Among the AI app builders, Lovable is the one with a first-party MCP server at the time of writing. If v0, Bolt or Replit ship one, nothing here needs to change.

  • BuildLovable
  • Code and deployGitHub, Sentry
  • DataSupabase, Postgres
  • MoneyStripe
  • Work trackingLinear, Notion
  • DesignFigma
  • CustomersZinta
Limits

What it will not do

It never sends anything

Zinta writes the opening message and stops. You send it from your own mailbox, so your domain reputation stays yours and a draft that got something wrong cannot reach a stranger without you reading it first.

It refuses to pad

When a hunt finds nothing that clears the bar, it says so. The filing tool rejects rows without a fetched source, a dated signal or a real address, and hands back the reason. A row you cannot check is a claim, and claims from a language model are how this whole category lost people’s trust.

Questions

Frequently asked

Do I need an API key?

Only for clients that read a config file, like Claude Code, Cursor or Codex. Claude's desktop and web apps take the URL on its own: they discover the authorization server, you sign in and approve once, and there is no key to copy or lose.

Whose model does the hunting?

Yours. Zinta supplies the brief and the memory of everything already seen, and the reasoning runs on the AI subscription you are already paying for. That is why this tier is $19 a month rather than $79: our marginal cost is a few database writes.

Does it send the emails?

No, and it will not. Zinta writes the opening message and stops. You send it from your own mailbox, so your domain reputation stays yours and a draft that got something wrong cannot reach a stranger without you reading it first.

What stops it inventing companies?

The filing tool refuses rows without a source URL that was actually fetched and whose page states the signal, a dated signal, and a published address rather than a guessed one. Rows that fail come back with the reason, so a hunt that finds nothing says so instead of padding.

Will it show me the same company twice?

No. Every hunt reads the pipeline before it searches, and that memory belongs to the account rather than the session. A company surfaced in one conversation will not come back in the next one, or next month.

Does it work outside Claude?

Yes. It is plain Streamable HTTP MCP with standard OAuth 2.1 and bearer keys, so any compliant client works. Nothing here is specific to one vendor.

$19 a month, on the AI you already pay for

A hundred companies a month, running on your own subscription. Managed plans, where we run it on a schedule and pay for the model, start at $79.

See plansHow it compares