You tell your agent to pull the latest tax messages from the government portal. It opens a browser tab, logs in with your saved credentials, navigates through three nested menus, and returns a tidy summary. The magic isn't that it can browse. The magic is that it leaves no trace, runs in your own Chrome, and produces a reusable script for next time. This is the quiet promise of Reduck, a browser-automation platform that treats the entire web as a surface for AI agents to discover, create, and execute workflows on sites that will never have an API [Perplexity Sonar Pro Brief, retrieved 2026].
Founded in 2025, Reduck is building for a world where agents are real workers, not just chatty assistants. Its core product, Reduck MCP, is a developer platform that sits between an AI agent and the web. An agent can explore a website, and Reduck generates a deterministic browser script from that exploration. That script is then exposed as a tool or endpoint via the Model Context Protocol (MCP), compatible with Claude, Codex, and any MCP-native application [LinkedIn, retrieved 2026]. The key architectural choice is local execution: scripts run through a Chrome extension in the user’s own browser, preserving logged-in sessions, cookies, and authentication states without ever exposing credentials to a third-party cloud [Station F, September 2026].
The Wedge of Determinism
Reduck’s bet is on reliability over generality. In the noisy, unstructured world of web scraping and automation, most solutions are brittle. They break when a site changes its layout, stumble on CAPTCHAs, and fail on complex login flows. Reduck’s approach is to let an agent learn a task once, in a real browser, and then codify that path into a repeatable script. The company claims this makes predefined workflows, like collecting invoices or conducting LinkedIn research, “three times faster and cheaper” than traditional step-by-step agent navigation [Tech.eu, October 2026].
The initial use cases are pragmatic, born from the unstructured corners of the internet where APIs fear to tread.
- Government portals. Automating the retrieval of tax notices or application statuses from legacy systems.
- Legacy business software. Accessing data from old dental practice portals or inventory systems.
- Social media research. Using official scripts to list followers or repost content on LinkedIn, Twitter, and Reddit while staying logged in [Reduck, retrieved 2026].
The product feels like a toolkit for a new class of integrator, someone who needs to bridge the gap between modern AI and the ancient, API-less web.
The Team and the Round
The founding trio of Daniel Huynh (CEO), Louis Abraham (CTO), and Raphaël Millet (COO) previously worked together at an unnamed company referred to as H Company [LinkedIn, retrieved 2026]. Huynh and Millet also share a background from Mithril Security, a privacy-focused startup, giving the team a pronounced orientation toward handling user data locally [LinkedIn, retrieved 2026]. This experience likely shaped the core architectural principle of running scripts in the user’s own browser, a significant privacy and security differentiator.
In October 2026, this team secured a $3 million pre-seed round led by European venture firm Daphni, with participation from Drysdale Ventures and Galion.exe [Tech.eu, October 2026]. The funding appears aimed at scaling the team and platform, as evidenced by open roles for a “Vibe Coder Intern” and a “Growth Intern” posted in Paris [Station F, September 2026].
| Role | Name | Notable Background |
|---|---|---|
| Co-founder & CEO | Daniel Huynh | Previously at H Company; co-founded Mithril Security [LinkedIn, retrieved 2026]. |
| Co-founder & CTO | Louis Abraham | Previously at H Company; led the technical pivot to web integrations [LinkedIn, June 2026]. |
| Co-founder & COO | Raphaël Millet | Previously at H Company; was Co-founder & COO at Mithril Security [Mithril Security Blog, retrieved 2026]. |
The Market’s Unstructured Frontier
Reduck is entering a market defined by a frustrating asymmetry: AI models are becoming astonishingly capable, but the digital world they need to interact with remains a patchwork of HTML forms and JavaScript. The company’s timing aligns with the rise of the MCP standard, championed by Anthropic, which is creating a common language for tools that AI agents can use. By building for MCP first, Reduck is plugging directly into the plumbing of the emerging agent ecosystem [Reduck, retrieved 2026].
Its competition is less a named startup and more a collection of alternatives: bespoke scraping scripts that require constant maintenance, cloud-based automation platforms that can’t handle logged-in sessions securely, and the manual labor the product seeks to replace. Reduck’s wedge is its combination of local execution for security and its focus on turning one-off agent explorations into reusable, deterministic assets.
Navigating the Inevitable Friction
For all its elegant promise, Reduck’s path is lined with technical and commercial friction. The most credible risk is the inherent brittleness of any system that interacts with the visual layer of the web. While deterministic scripts are more robust than dynamic scraping, a major website redesign could still break a workflow, requiring the agent to re-learn the task. The company’s answer likely lies in the sophistication of its script-generation and its library of official, maintained scripts for major platforms like LinkedIn.
Another challenge is scaling the developer mindshare. MCP is gaining traction but is not yet ubiquitous. Reduck’s growth is tied to the adoption of agentic workflows themselves. Its pricing model, with a free tier offering 100 script runs per month and a $20 plan for 1,000 runs, is designed to attract early developers and tinkerers [Reduck, retrieved 2026]. The bet is that these users will build the workflows that prove the platform’s enterprise value.
The Next Twelve Months
The fresh capital provides an 18-month runway to move from a clever developer tool to a must-have platform. The key milestones to watch will be less about vanity metrics and more about ecosystem depth.
- Script library growth. Expanding beyond social media to official scripts for major e-commerce, SaaS, and government sites.
- Enterprise adoption. Signing its first paid deals with companies using Reduck to automate internal, legacy-software workflows.
- MCP ecosystem expansion. Becoming a default, recommended tool within the Claude desktop and other agent hubs.
The cultural question Reduck is implicitly answering is not about whether agents will use the web, but how they will claim it. Will it be through fragile, cloud-based puppetry that treats our browsers as remote terminals? Or through localized, user-owned tools that treat the browser as a partner? Reduck is betting on the latter, building a platform where the agent’s playground is, quite literally, your own tab.
Sources
- [Perplexity Sonar Pro Brief, retrieved 2026] Reduck MCP product description | https://reduck.ai/
- [LinkedIn, retrieved 2026] Reduck AI company page and product details
- [Station F, September 2026] Job listing describing product architecture and funding | https://jobs.stationf.co/companies/conception-dev/jobs/stage-vibe-coder_paris
- [Tech.eu, October 2026] Reduck secures $3M pre-seed round | https://tech.eu/2026/10/22/reduck-secures-3m-to-make-web-automation-faster-for-ai-agents
- [LinkedIn, June 2026] Louis Abraham launch post on company pivot | https://www.linkedin.com/posts/louis-abraham_for-months-we-iterated-in-stealth-we-kept-activity-7467100161110917120-xrg3
- [Mithril Security Blog, retrieved 2026] Raphaël Millet background | https://mithrilsecurity.io
- [Reduck, retrieved 2026] Pricing page and script library | https://reduck.ai/pricing