Reduck raises $3M to build the web fast lanes for AI agents
Today we’re announcing a $3M pre-seed round led by Daphni, with participation from Drysdale Ventures and Galion.exe to help us on our mission to connect agents everywhere.
Why now
Agents are supposed to revolutionize our lives, but in 2026 they still barely operate where we do, as many sites do not provide an MCP or API for agents to use them.
Key platforms such as Amazon or LinkedIn want to keep a direct relationship with their customers, especially when their business model depends on ads or paid services for automated access. Amazon illustrated this on September 20 by blocking Muse, Meta’s personal agent, thus slowing down AI agent adoption.
Beyond B2C scenarios, many B2B agent use cases have not been deployed, as most internal tools do not currently offer agent-native APIs.
To circumvent this lack of API, agents are now trained and able to click and type in a browser to automate sites that provide no programmatic access. However, this is far from being ready for complex tasks, as it is:
- Slow. The agent clicks for minutes on a job that should take seconds. People wait, and the quick back-and-forth that makes an agent useful stops working.
- Expensive. Every screenshot goes into the context. On a subscription, you hit rate limits. On the API, you pay for every token.
- Unreliable. The agent takes a different path each time, and its results get worse as its context fills up, so hallucination rates increase.
Introducing Reduck MCP
That’s why we built Reduck MCP: the easiest way for an agent to integrate and automate any site you use.
Reduck MCP allows agents to discover, run and create browser automation scripts that serve as tools. By doing so, agents no longer need to learn and click on a website on the fly and simply get results instantly.
This enables agents to be on average 3x faster and cheaper on a wide array of tasks from downloading 20 invoices from 5 different vendors, to doing market research on LinkedIn. You can find a more detailed benchmark in our article covering our benchmark Reduck Voyager.
Customer success stories
Thanks to Reduck MCP, our customers have been able to automate repetitive browser tasks to save hours:
“Fetching my SaaS invoices used to take five hours a month. Reduck does it for me in half an hour.”
Marie Brayer, Partner & Founder of Mox Venture VC
“We went from an afternoon of copy-pasting Sales Navigator profiles to a two-minute export with Reduck.”
Tim Liu, Head of Product at BentoML (acquired by Modular)
“It used to take hours of manual entry to fill prospects’ accounts with real data before a B2B enterprise demo. Reduck does it in minutes.”
Alexandre Grux, CPO at Hyperlex
You can find more about what you can do with Reduck by exploring our use cases.
Meet the team
Reduck is a French-American startup built by three experienced repeat founders: Daniel Huynh (CEO), Louis Abraham (CTO) and Raphaël Millet (COO).
Daniel and Raphaël co-founded Mithril Security, which was acquired by H Company (founded by DeepMind engineers, raised $220M). Before the acquisition, Mithril Security launched LaVague, the first open-source AI web agent (6k GitHub stars), and deployed privacy-by-design LLMs on hardware-based secure enclaves, supported by an OpenAI grant.
Louis is a leading AI engineer. He co-founded Secrecy, an end-to-end encryption platform built on patented cryptographic protocols, worked with Stanford on biomedical machine learning research, and was an ICPC finalist. ICPC is the world’s most prestigious competitive programming contest. He met Daniel and Raphaël at H Company, where he was a core team researcher and led the agent production team.
With the funding, we’ve grown the team to eight people with complementary profiles across AI engineering, full stack and infrastructure.
What’s next
We are focusing on continuously improving the reliability and number of scripts of our official library (currently 1,200+ ready-to-run scripts are available across 250+ websites). We have been iterating on our agent swarm that can create and fix scripts autonomously at scale.
We are also preparing a developer offer, so that teams can build custom agent harnesses that are connected to environments without API, such as specialized portals in insurance or healthcare, or custom internal CRMs.
Finally, we are looking at deploying vertically integrated apps that leverage our scripts to automate key knowledge work, beginning with GEO.