Agent Field Guide @AgentGuideHQ

Best MCP servers by use case: agentfieldguide.net/guides/best-mc…

Your AI agent can't see your repo. MCP fixes that. What the Model Context Protocol is, in 45 seconds 👇 Which MCP server do you use most?

@choopyplug1 Two that pay off fast: a PreToolUse hook that exits with code 2 blocks the command and sends the reason back to Claude, so it picks a safer route. And give the review subagent a tools list of only Read, Grep and Glob, so it can't edit even by accident.

Side-by-side page: agentfieldguide.net/compare/langgr…

LangGraph vs CrewAI, in one card. LangGraph: low-level graph orchestration with persistence and human-in-the-loop. Python + TypeScript, MIT. CrewAI: role-playing multi-agent "crews" with tasks and processes. Python, open source + paid platform. Graph or crew?

Figma Dev Mode MCP: design context for your agent (paid Figma plans) Linear MCP: issues, projects and comments (free with Linear) Which one would you add first? Full list with licenses and links: agentfieldguide.net/guides/best-mc…

Supabase MCP: tables, queries and project config Postgres MCP Pro: queries, index tuning and health checks Sentry MCP: issues, errors and performance data Firecrawl MCP: web scraping, crawling and search

10 MCP servers that make Cursor's agent more useful. One line each on what it actually does 👇

@stretchcloud Agents as board members is the right shape. Even on a regular tracker the same idea helps: give each agent its own account and a token scoped to its project, so the history shows which agent changed what and a bad run can't touch other boards.

@gregpr07 Two things make this stick: have the agent save the query or steps it ran next to the chart, so next week is a rerun instead of a fresh browse. And give the browser session a read-only login. It's driving a real account, so it shouldn't be able to change anything.

Full table with what each one does: agentfieldguide.net/category/frame…

18 agent frameworks in one table. LangGraph, CrewAI, AutoGen, OpenAI Agents SDK, Pydantic AI, smolagents, Mastra, Google ADK, DSPy, Letta and more. Python-only: CrewAI, Pydantic AI, smolagents, DSPy TypeScript-native: Mastra, Vercel AI SDK Which one are you shipping with?

Solid list. One thing worth adding under MCP: keep it to 2–3 servers. Every server loads its tools into Claude Code's context, and too many makes it pick the wrong one. A good starter set for this stack: Context7 for version-specific docs, Supabase MCP for the DB, Playwright MCP so it can check its own UI.

Which agent framework, MCP server or coding agent should you use? Agent Field Guide is live: an independent directory of AI agent tools. Every entry links to the official site or repo, and sponsored spots are always labeled. What should we add next? agentfieldguide.net

Open-weight multimodal models landing inside coding agents is the pattern to watch. @cline making Nemotron 3.5 Super VL free to try is a good test: does vision+docs in the agent actually cut tool calls, or just burn more tokens?

Cline @cline

7 days ago

Nemotron 3.5 Super VL is available for free in Cline! 120B param open-weight model with 12B active, delivers 5x faster output than Nemotron 3 Ultra, and has native support for images, videos, and documents. Great work @NVIDIAAI ❤️

6 10 141 7K 17

@AverageAiBro Shell dialect mismatch is such an underrated agent failure mode on Windows. Shipping the translator as an MCP server is clever — the agent keeps its muscle memory and the host stays native.

@pydantic Prod traces inside the coding agent is the right idea. Pairing Logfire MCP with something like Sentry MCP for errors closes the loop between "write code" and "see what broke in prod."

@cline @NVIDIAAI Open-weight VL + native tool use inside a coding agent is a big unlock — especially when the active params stay small enough to feel snappy. Nice work shipping this @NVIDIAAI.

@LangChain @NVIDIAAI 87% on a behavioral suite for Deep Agents is the number that matters more than the param count. Vision + tool use in one open-weight checkpoint is exactly what harness builders have been waiting for.