London-based Archestra has secured €2.8 million in an oversubscribed pre-seed, closing the deal in less than two weeks.
The startup promises to solve one of AI’s biggest headaches: what happens when autonomous agents gain too much access—and start making decisions they shouldn’t.
Key Takeaways
- Archestra raised €2.8M pre-seed in under two weeks.
- Investors include Concept Ventures, Zero Prime, RTP Global.
- Startup builds guardrails for Anthropic’s MCP framework.
- Founders previously sold startups to Grafana and Elastic.
- Enterprise-ready AI tools promise productivity without chaos.
Archestra, a London startup, raised €2.8 million to build security guardrails for AI agents using Anthropic’s Model Context Protocol. The open-source platform prevents “rogue” behavior—like data leaks or file deletion—while allowing companies to safely connect AI to systems such as email, Slack, and HR platforms.
A funding story with bigger implications
The AI gold rush isn’t just about building smarter models—it’s about controlling them. London-based Archestra has raised €2.8 million in a lightning-fast, oversubscribed pre-seed round, aiming to solve a risk that keeps CIOs awake at night: what happens when AI agents go “rogue.”
The round, closed in less than two weeks, was led by Concept Ventures, with backing from Zero Prime Ventures, Celero Ventures, RTP Global, and Aloniq. Notable angels include Maxim Konovalov (Nginx), Stephen Whitworth (incident.io), and Luke Harries (ElevenLabs).
Why this startup matters
AI agents powered by large language models can draft emails, edit documents, and even fetch data from internal systems. But without checks, the same agent could access salary spreadsheets—or delete files—without approval.
That risk is real enough to scare off enterprises. Archestra claims it has the fix: an open-source “MCP orchestrator” that adds permissions, data management, and security layers around Anthropic’s recently unveiled Model Context Protocol (MCP).
“We’re building the security-first solution that will change this,” said CEO Matvey Kukuy, who previously sold startups to Grafana and Elastic.
Investors see an “API moment” for AI
Backers argue the technology could become as foundational as APIs were for the internet.
“Just as APIs became the building blocks for web infrastructure, MCPs are emerging as the connective tissue for enterprise AI,” said Ariel Rahamim, Principal at Concept Ventures. “Archestra is building the infrastructure layer this ecosystem needs.”
The people behind it
Founded in 2025, Archestra is led by Kukuy and Ildar Iskhakov, longtime friends and serial entrepreneurs. They are joined by Joey Orlando, an ex-Grafana engineer. Their track record includes Amixr, sold to Grafana in 2021, and Keep, sold to Elastic soon after.
That pedigree gives investors confidence the team can execute in an increasingly crowded AI safety market.
Why it matters
Enterprises can’t afford rogue AI agents. Archestra’s platform offers a middle path: keep productivity gains but build in guardrails. In a market where the integration-platform-as-a-service (iPaaS) sector is forecast to exceed €14.5 billion by 2028, that’s a massive opportunity.
Impact
For everyday employees, this means AI assistants could soon become part of daily work—helping write reports, manage calendars, or handle HR requests—without exposing personal or sensitive company data. Safer AI tools don’t just help IT teams; they help everyone in the office.
Numbers to Watch
- €2.8M raised in oversubscribed pre-seed
- <2 weeks to close the round
- €14.5B projected iPaaS market size by 2028
What’s Next
- Roll out enterprise beta for early adopters.
- Expand team in London and abroad.
- Add user-friendly controls for non-technical staff.
- Target regulated industries where AI adoption lags.
Conclusion
Archestra’s fast raise shows investors aren’t just chasing smarter AI—they’re betting on safer AI. If the startup delivers, it could become the default guardrail for enterprises experimenting with autonomous agents.
Tomorrow’s question: will “safety-first” become the winning edge in enterprise AI—or will speed and power win out?