toolweave is two things that fit together.
First, it's a single MCP gateway: one endpoint that gives Claude (or any MCP client) access to around a hundred tools — image and video generation, financial data, a crypto exchange, web scraping, a real browser, email, DevOps, and your own custom APIs. You connect once and Claude can use all of them.
Second, it's a team platform: a way to turn those tools into virtual employees who own tasks and do the work on their own schedule, without you in the loop. The gateway is the hands; the team is what puts them to use.
Most people meet the gateway first and grow into the team.
The gateway: one connection, many tools
Plugging tools into an AI assistant one by one is tedious, and every new integration is another thing to wire up and maintain. toolweave collapses that into a single MCP endpoint. Behind it sit tool families you can mix and match:
- AI generation — text, images, video, speech, music (OpenAI, Gemini, fal, Suno, Claude).
- Financial data — quotes and fundamentals from Alpha Vantage, Finnhub, Twelve Data, and SEC filings.
- Crypto — live market data and trading on Kraken.
- Web — search and scrape pages into clean markdown (Firecrawl).
- Browser — drive a real browser: click, type, extract (Stagehand).
- Email — read a mailbox and send mail over IMAP/SMTP.
- DevOps — GitHub, Vercel, Railway.
- Custom connectors — point toolweave at any HTTP API's docs and get working tools, no code.
You bring your own keys for the providers you use, so the calls bill to your accounts, not through a markup. The point isn't a pile of integrations — it's that they're all one connection away.
The team: tools that work without you
Tools you call by hand are useful. Tools that work while you sleep are a different thing entirely. That's the second half of toolweave: virtual team members.
A member is a named employee with a role, standing context, a set of skills (the tool families above, whitelisted), and tasks they own. When a task runs, the model acts in character as that member — it decides and signs its work as them, not as a generic assistant.
And tasks don't wait for you. Each one picks how it starts:
- on a schedule (a cron time),
- on an incoming email (it reacts to new mail in the member's own mailbox),
- on a webhook (an external system POSTs and the body becomes the assignment),
- or by hand (a one-off instruction).
So a member can review the markets every morning, triage support mail as it arrives, or react to a form submission — and deliver the result to your inbox, in a visual style you chose for them.
What keeps it safe and sane
Autonomy without guardrails is a liability, so three things hold it together:
- Whitelists. A member can only ever touch the skills you ticked. Nothing more is even visible to it.
- Memory. Give a member the memory skill and it remembers across runs — what worked, what a contact prefers, how a process really goes. It improves instead of restarting from zero each time.
- The approval gate. For anything public or irreversible — publishing, sending outbound mail, deleting — a member doesn't act. It proposes, the action lands in your queue, and it happens only when you approve. Autonomy for the safe parts, a sign-off for the rest.
Every run is also recorded step by step — which tools it called, what they returned, where it ended — so you can always see what actually happened, not just the final answer.
Who it's for
toolweave is built for solo founders and small teams who need leverage, not headcount. Instead of stitching together a dozen SaaS automations, you hire a few virtual employees: a market analyst, a support rep with its own mailbox, a watchdog that probes your endpoints each morning, a researcher, an editor. They can even read each other's activity, so a coordinator can see what the rest of the team did.
The shortest way to say it: toolweave gives Claude a hundred tools, then lets you hire people to use them. Start with one connection, grow into a team.
Want the hands-on version? Read Build your first AI employee.