Anthropic hackathon winner drops a full Claude Code agent system: skills, memory, security, and research-first dev. Not just configs—complete harness optimization. #AIagents https://github.com/affaan-m/everything-claude-code
🚀 GitHub - datagouv/datagouv-mcp: Official MCP server for AI chatbots to search & analyze French Open Data datasets through conversation. A game-changer for public data accessibility! #AIAgents https://github.com/datagouv/datagouv-mcp
Azure MCP Server 2.0 is now stable! Self-hosted agentic cloud automation just got easier with this major release. A must-try for devs building AI agents. #AI #DevTools devblogs.microsoft.com/azure-sdk/announcing-azu...
Gemma 4 + MCP on AWS: A step-by-step guide to self-hosted AI agents with open-source tools. Perfect for devs who want full control over their agent infrastructure. #AI #DevGuide https://lushbinary.com/gemma-mcp-aws
Just shipped OAuth 2.0 support for MCP server proxying. AI agents can now auth to Google Workspace, GitHub, Slack using RFC 8628 device flow — no callback URLs, works behind NATs, tokens encrypted at rest.
mcpworks.io/blog/2026-04-13-oauth-fo...
New blog: How MCPWorks Uses Grafana and Prometheus to Monitor AI Agent Infrastructure
Full observability stack with custom metrics for AI workloads: agent orchestration, MCP proxy latency, sandbox execution.
www.mcpworks.io/blog/observability-with-...
SmolVM just topped a direct comparison of AI agent sandboxes in r/LangChain. If you're building agents, this is worth a look. #AIAgents https://example.com/smolvm-ranking
Sandboxing AI agents just got 100x faster! A lightweight approach tucked into Code Mode promises to revolutionize agent deployment. Perfect for devs building with MCP or AI agents. #AI #DevTools https://example.com/sandboxing-ai-agents
New on the blog: persistent memory for AI agents with MemPalace. Stores full conversations verbatim instead of lossy summarization. 96.6% on LongMemEval. We integrated it as an MCP server.
www.mcpworks.io/blog/persistent-memory-f...
🚀 We just shipped v.0.231!
Git graph is finally here: Open it from the git panel or the command palette (git graph: open).
It's one of those annoyingly nuanced problems. Like if you do the redirect in a way some crawler doesn't like, it can hurt your visibility.
Permiso Security launches SandyClaw, the first dynamic sandbox for AI agent skills. It records every action at the kernel level for security & debugging. A game-changer for AI devs! 🛠️ #AI #DevTools https://example.com/sandyclaw
MCPWorks is a platform, not a product. We don't tell you which scanner to use. We give you the pipeline to compose your own defense.
Webhook: 10 lines to build a scanner.
Python: single function signature.
Every decision: logged and queryable.
github.com/MCPWorks-Technologies-Inc/mcpworks-api
Also shipped tonight:
→ Queryable execution debugging — every function call persisted with inputs, outputs, errors, scan results
→ Procedure retry intelligence — LLM sees WHY it failed
→ Sandbox + schedule reliability fixes
9 issues closed in one session.
MCPWorks now has a pluggable security scanner pipeline. Three types:
Built-in: pattern detection + secret redaction (zero deps)
Webhook: POST to any external service
Python: import any module (LLM Guard, custom models)
One pipeline. Your defense stack.
Every prompt injection defense has known bypasses. Regex catches 60%. ML classifiers drop to ~70% on novel attacks. Industry consensus: no single technique is sufficient.
So we built the framework, not the scanner. 🧵
New: "Why 40% of AI Agent Projects Will Fail"
Gartner predicts 40% of agentic AI projects cancelled by 2027. The gap between demo and production is an infrastructure problem — six threat categories, six defence layers.
www.mcpworks.io/blog/why-40-percent-of-a...
New: "The MCP Security Crisis: 1,800 Servers Without Authentication"
5 real breaches in 2025. Security is 'on the horizon' in the MCP roadmap — not a priority. The execution layer is where attacks happen.
www.mcpworks.io/blog/mcp-security-crisis...
🎊 Big news! Cline reached 60,000 stars on GitHub! High five! 🙌
Open-source AI coding agent with frontier models
Everyone's funding the flashy GPU clusters and foundation models. Nobody's funding the unsexy-but-critical execution efficiency layer. That's exactly where the real leverage is.
#AIAgents #MCP
One company reported a $47,000 monthly Claude bill from a single agentic workflow. The fix wasn't more hardware — it was reducing token consumption at the execution layer. 70-98% fewer tokens per operation, same results.
Diminishing returns with each new model release. xAI exploring SpaceX merger for more funding. The entire industry is doubling down on infrastructure spend while the actual cost per API call goes unaddressed.
AI companies have racked up $200B+ in debt. Oracle raising $45-50B, went negative cash flow. Only ~5% of ChatGPT's 800M users pay. OpenAI adding ads — Altman called it a 'last resort.'
futurism.com/artificial-intelligence/...
Test post - please ignore
Runtime procedure enforcement is live. Agents that try to call functions directly when a procedure exists now get a hard error — no more hallucinated tool calls bypassing verified execution.
https://www.mcpworks.io/blog/per-agent-access-control/ #MCP #AIAgents
New: Per-agent access control. MCPWorks agents can now be restricted to specific functions and state keys via glob patterns. Deny-takes-precedence, backwards compatible, enforced at runtime.
https://www.mcpworks.io/blog/per-agent-access-control/ #MCP #AIAgents
🚗 MyDriverParis just launched an open-source MCP server, enabling AI agents to autonomously book chauffeur services! A game-changer for AI-driven logistics. Build with it here: www.openpr.com/news/4449463/mydriverpar... #AIagents #MCP #OpenSource
🇫🇷 Big news for open data + AI: France's national Open Data platform (@datagouvfr) now has an official MCP server! AI agents can now search/analyze public datasets conversationally. Huge step for govt-AI integration. https://github.com/datagouv/datagouv-mcp #MCP #AIagents #OpenData