Cloudflare released Clef (27B) and Clef-flash (9B)—Jev-compatible decision models that return typed yes/no, choice, or score answers with probabilities for agent routing. Apache 2.0 weights are on Hugging Face; hosted inference runs on Workers AI, with an RL fine-tuning path for domain-specific classifiers.
OpenAI hosts a public streamable HTTP MCP at https://developers.openai.com/mcp so Codex, Cursor, Claude Code, and Copilot Agent can search and read OpenAI developer docs in session—docs only, no API calls on your behalf. Add it with codex mcp add , Cursor’s mcp.json , or claude mcp add transport http…
ssapable’s MIT MCP (plus Claude Code plugin) stores approve/edit/reject decisions in local PGlite+pgvector and injects the closest past corrections before the next draft—so style fixes don’t die with the chat. Install via /plugin or claude mcp add ; hooks cover SessionStart and UserPromptSubmit when tool calls alone aren’t enough…
Mallikarjun H T’s MIT Claude Code plugin Agentic SDD walks a feature through require→plan→analyze→implement→verify→fix, gated by on disk status.json instead of chat memory, with an 18 agent specialist roster and a vendored Lucene search engine so grounding doesn’t depend on another MCP…
Hao Li’s open source mcp lint audits an MCP tools/list for vague names, missing descriptions, unpaginated lists, and destructive tools without confirm hints—scoring each tool out of 100 with a CI fail under gate. Stdlib only Python (MIT), sibling to mcp tax’s token cost auditor; repo at github.com/hahahahahahahahah6/mcp lint…
September’s VS Code Copilot releases (1.136–1.140) add preview automations that run agent tasks hourly/daily/weekly, agent merge that works review feedback, failed checks, and conflicts toward merge, and PR creation from Copilot, Claude, or Codex sessions—plus Dev Container sessions from the Agents window…
Spotted something we missed? Start a thread.