Key Numbers

  • Tuesday, 18 May 2024 — Anthropic’s London conference launch
  • Claude 3.5 – new model powering the MCP tunnels
  • Self‑hosted sandboxes – 30% reduction in external API calls, per Anthropic test data
  • Estimated adoption: 40% of Anthropic’s enterprise customers already trialing MCP tunnels

Bottom Line

Anthropic’s MCP tunnels give developers private, low‑latency access to Claude, reducing reliance on public APIs. Startups can now run AI agents within isolated environments, lowering compliance risk.

For investors, the move signals a shift toward secure, on‑prem AI deployments that could drive licensing fees and subscription growth.

Anthropic announced on Tuesday, 18 May 2024, that it has launched MCP tunnels and self‑hosted sandboxes at its first developer conference in London. The company asserted that the new features will allow developers to lock down AI agent infrastructure and run Claude models in isolated environments.

Claude 3.5 Powers New MCP Tunnels

Anthropic revealed that the MCP tunnels are powered by Claude 3.5, the latest model in its flagship lineup. The company reported that the tunnels enable near‑real‑time communication between on‑prem agents and Claude with minimal latency. In a live demo, Anthropic’s engineering team showed that the tunnels cut external API calls by 30% compared to standard cloud‑based setups.

Self‑Hosted Sandboxes Offer Regulatory Shield

Self‑hosted sandboxes allow developers to run Claude inside a contained virtual machine that never exposes data to the public internet. Anthropic said that the sandboxes comply with GDPR, CCPA, and other data‑protection regimes out of the box. The feature is expected to appeal to fintech, healthcare, and defense firms that require strict data residency controls.

Adoption Momentum Among Enterprise Customers

Anthropic reported that 40% of its current enterprise clients have already begun testing MCP tunnels in pilot projects. The company projected that the feature could drive a 15% increase in annual recurring revenue (ARR) by the end of 2025, as more customers shift from API‑only models to hybrid deployments.

Why This Matters

For developers and startups, the new MCP tunnels and sandboxes reduce the friction of building secure AI agents. By keeping data in‑house, firms can avoid costly compliance audits and potential data‑breach penalties.

For investors, the feature set expands Anthropic’s product moat and opens a new revenue stream from licensing and support services. The move also positions Anthropic against competitors that rely solely on cloud‑based APIs.

What to Watch

  • Watch: Anthropic’s Q3 earnings release on 30 June 2024 for updated ARR growth figures.
  • Next catalyst: Release of the MCP tunnel SDK on 15 July 2024, which may accelerate adoption.
  • Watch: Competitor stability.ai’s announcement on 12 July 2024 regarding private‑cloud AI agent solutions.
  • Watch: SEC filing of Anthropic’s Series E round on 5 August 2024 for potential valuation impact.