FIND-20260326-003 · 2026-03-26 · Innovation Veille
OpenClaw v2026.3 — Model-Agnostic AI Agent Platform with Claude Code + MCP Support (March 2026 Update)
tool
MEDIUM
Alex Finn highlighted the March 2026 OpenClaw update as a major release. OpenClaw is a free, open-source autonomous AI agent (199K GitHub stars) that runs locally across messaging platforms (Slack, Teams, WhatsApp, Telegram, Discord). The March 2026 update adds: support for Claude Code and Cursor MCPs, Claude Code config import, Exa web-search integration, a ClawHub plugin marketplace, and support for multiple AI providers (Claude, GPT-4o, Gemini, DeepSeek, local Ollama). New release v2026.3.24 also adds Docker/Podman container execution flags (--container), Gateway/OpenAI-compatible endpoints (/v1/models, /v1/embeddings), Slack rich-reply parity, and hardened sandbox security. IMPORTANT SECURITY NOTE: A prior critical RCE vulnerability CVE-2026-25253 (CVSS 8.8) affected all versions before 2026.1.29 — exposed ~40K instances to one-click RCE via auth token exfiltration through a malicious gatewayUrl parameter. Patched in v2026.1.29 (Jan 2026). Current version 2026.3.24 is patched. Separately, ~12% of ClawHub community skills were found to contain malicious code — supply chain risk for any skill installations.
Source
https://x.com/alexfinn/status/2036677722400035067
ODS Impact
LOW direct impact on current ODS stack. OpenClaw is a general-purpose AI agent (life/work automation) rather than a developer infrastructure tool. However, two areas are worth noting: (1) The Claude Code + MCP integration could be useful for ODS agents running on messaging platforms — if ODS ever needs AI-driven Slack/Teams bots with tool-calling, OpenClaw's architecture is a reference. (2) The Docker container execution flag (--container) could complement ODS's Coolify-based deployments if OpenClaw were ever used as a workflow automation layer. Not recommended for adoption in current P0-P1 pipeline work.
Security Review
License: MIT | Maintenance: ACTIVE | Risk: HIGH | Recommendation: USE_WITH_CAUTION
Tags
ai-agent
claude-code
mcp
slack
automation
open-source
security-cve
multi-model