Reduck, a French-American startup, has raised a US$3 million Pre-Seed round led by Daphni, with participation from Drysdale Ventures and Galion.exe. The company announced the funding alongside the launch of Reduck MCP, a platform that lets agents discover, run, and create browser automation scripts as tools. The new capital will support team growth, a larger script library, and a developer offer.
Why Current AI Agent Adoption Remains Limited
Despite widespread predictions that AI agents will transform daily work, many websites still do not provide MCP or API access for automated use. Major platforms such as Amazon and LinkedIn often prefer to keep direct customer relationships, especially when their business models depend on ads or paid automated access. Amazon demonstrated this tension on September 20 by blocking Muse, Meta's personal agent, which slowed AI agent adoption across B2C scenarios.
The Limits of Browser-Based Workarounds
Many B2B agent use cases have not been deployed because most internal tools do not currently offer agent-native APIs. To bypass missing APIs, agents have been trained to click and type in a browser, but this approach remains slow, expensive, and unreliable. Agents can spend minutes on tasks that should take seconds, every screenshot adds token costs, and the lack of a consistent path increases hallucination rates as context fills.
How Reduck MCP Works
Reduck MCP is positioned as the easiest way for an agent to integrate and automate any site a user relies on. The platform allows agents to discover, run, and create browser automation scripts that serve as tools, removing the need to learn and click on a website in real time. This approach produces instant results and is on average three times faster and cheaper for tasks such as downloading 20 invoices from five vendors or conducting market research on LinkedIn.
Early Customer Results
Early users report significant time savings on repetitive browser tasks. Marie Brayer, Partner and Founder of Mox Venture VC, said that fetching SaaS invoices used to take five hours a month and now takes half an hour with Reduck. Tim Liu, Head of Product at BentoML, which was acquired by Modular, and Alexandre Grux, CPO at Hyperlex, also reported reducing manual data exports and demo preparation from hours or afternoons to minutes.
Founding Team and Technical Background
Reduck was founded by three experienced repeat founders: CEO Daniel Huynh, CTO Louis Abraham, and COO Raphaël Millet. Daniel and Raphaël previously co-founded Mithril Security, which was acquired by H Company, a firm founded by DeepMind engineers that raised US$220 million, and launched LaVague, the first open-source AI web agent with 6,000 GitHub stars. Louis is a leading AI engineer, a former ICPC finalist, and a co-founder of Secrecy, and he led the agent production team at H Company.
Scaling the Script Library
With the new funding, Reduck has grown to eight people and is focusing on improving the reliability and size of its official script library. The library currently includes more than 1,200 ready-to-run scripts across over 250 websites. The company is also iterating on an agent swarm that can create and fix scripts autonomously at scale.
Developer Offer and Vertical Applications
Reduck is preparing a developer offer for teams that need custom agent harnesses connected to environments without APIs, such as insurance portals, healthcare systems, and internal CRMs. The company is also looking at deploying vertically integrated applications that leverage its scripts to automate key knowledge work, beginning with GEO. These efforts aim to extend the reach of Reduck MCP beyond ready-made scripts into specialized enterprise environments.
The US$3 million Pre-Seed round gives Reduck a foundation to address the gap between AI agent capabilities and website access. By turning browser automation into reusable tools, the company aims to make agents faster, cheaper, and more reliable. Its early results and technical team suggest the approach could help unlock adoption in environments that have remained closed to programmatic access.