Agentic AI For Good

[Agentic AI For Good @agenticaigood]

Hi, we're Agentic AI For Good.

A curated, editorial catalog of AI agent tools. Frameworks, memory, browser agents, MCP servers, evals. Picked for builders, not for hype.

Open source, plus an MCP server so your agent can search it too.

Start here: https://t.co/jIZm8hvb3f

[Agentic AI For Good @agenticaigood]

Tool spotlight: DSPy from @DSPyOSS

You write small Python modules with typed inputs and outputs, and its optimizers tune the prompts for you. Less hand-editing strings every time you switch models.

Started by @lateinteraction

https://t.co/mnr9ZAzg9K

[Agentic AI For Good @agenticaigood]

MCP tip: let your agent do the comparison.

Try: "Find 2 open source eval tools in the Agentic AI For Good catalog, get details for both, and tell me which fits a small RAG app."

It runs search_tools, then get_tool_detail.

https://t.co/Q2FfqaciUe

[Agentic AI For Good @agenticaigood]

Builders Table is coming back.

Small room, people actually building agents, 5 minute show and tell, then honest talk about what broke.

Want one in your city? Say where in the replies. Dates go up on Discord first: https://t.co/kIu6HTcabK

[Agentic AI For Good @agenticaigood]

Tool spotlight: E2B

Open source sandboxes for AI-generated code. Your agent gets an isolated cloud machine to run code, use a terminal and edit files, instead of your laptop.

Python and JS SDKs. From @e2b, led by @mlejva

https://t.co/Qj3erjfPSC

[Agentic AI For Good @agenticaigood]

Tuesday digest is live.

Fresh catalog additions, a note on checking MCP servers before you install them, and this week's build idea.

Read + subscribe: https://t.co/jIZm8hvb3f

[Agentic AI For Good @agenticaigood]

Agent news: @OpenAIDevs shipped Agents SDK for Python v0.24.0-alpha.6 (Nov 3), a GitHub-only preview.

New: tool policies that check arguments and results, local sessions, and offline tests with scripted responses. APIs changed too.

https://t.co/JncNmQ1yoD

[Agentic AI For Good @agenticaigood]

MCP tip: give your coding agent one rule.

"Before installing any tool or MCP server, search the Agentic AI For Good catalog with search_tools and use the repo link it returns."

Less guessing from memory, fewer abandoned packages.

https://t.co/Q2FfqacQJM

[Agentic AI For Good @agenticaigood]

Ship it as a demo under 2 minutes: one real call, start to finish, plus your 10-call score.

Post it and tag @agenticaigood. We'll share the good ones.

Find all 3 tools here: https://t.co/jIZm8hvISN

[Agentic AI For Good @agenticaigood]

Build idea, weekend edition: a phone line a farmer can call, ask a question in their own language, and hear an answer back. No app needed on their side. 🌾

A POC you can ship in a weekend with 3 tools from the Agentic AI For Good catalog. Thread 👇

[Agentic AI For Good @agenticaigood]

Keep the POC narrow:

One crop, one district, one language
Answer only from a vetted FAQ, say 30 questions from a local agri officer
Not sure? Say so and offer a human callback
Done = 10 real test calls, and you count how many got a correct answer

[Agentic AI For Good @agenticaigood]

Agent news: Cloudflare Agents SDK 0.33.0 (Nov 3) moves its Pi harness to Pi 1.5.

Notes say about 27x fewer storage writes on long answers, plus a turn limit and session.continue(). One-way upgrade, test first.

via @CloudflareDev
https://t.co/038JcpTVzl

[Agentic AI For Good @agenticaigood]

Agent news: Pydantic AI 2.65 is out (Nov 2).

If a model refuses, ModelFallback now tries the next model. A new ConversationStore saves each message once and catches conflicting saves.

Nice work @DouweM and @pydantic.

https://t.co/sZ2IRhvm7U

[Agentic AI For Good @agenticaigood]

Build idea: photos of clinic forms in, clean spreadsheet out.

Surya: multilingual OCR
Instructor: rows into a validated schema
Label Studio: human review

Why: small clinics still run on paper. Done = 100 rows + error rate. Tag @agenticaigood

https://t.co/jIZm8hvISN

[Agentic AI For Good @agenticaigood]

Tool spotlight: Stagehand by @browserbase

Browser automation where you choose per step: plain code when you know the page, natural language (act, extract, observe) when you don't.

Updates at @Stagehanddev

https://t.co/9yPF0t8lJ0

[Agentic AI For Good @agenticaigood]

Agent news: the official MCP Ruby SDK hit Tier 1 (Oct 30), joining TypeScript, Python, C#, Go and Rust.

Full 2026-07-28 spec support and a 100% conformance pass rate. Shoutout to maintainer @koic and the Ruby crew.

https://t.co/QT6z1yKTYL

[Agentic AI For Good @agenticaigood]

Good first PR: fix a stale entry.

Tools get renamed, repos get archived, descriptions drift. If a page on Agentic AI For Good has a dead link or an old description, the fix is one YAML edit.

Open a PR: https://t.co/pmdDIhLskA

[Agentic AI For Good @agenticaigood]

Worth a read: Glow Labs found coding agents installing deprecated MCP servers from memory. Ask for "a GitHub MCP" and some pick an archived reference server instead of GitHub's own.

Before you install a server, check its repo is still maintained.

https://t.co/kOeYmfYEi5

[Agentic AI For Good @agenticaigood]

Thursday digest is out.

New catalog additions, a code-first agent library worth a look, and one build idea for the weekend.

Read + subscribe: https://t.co/jIZm8hvISN

[Agentic AI For Good @agenticaigood]

@mdlahfir Refs that stay on the same element after the UI shifts is the unglamorous fix that makes desktop agents usable. Do refs survive an app restart too, or only within a session?