Compresses tool outputs, logs, files, and RAG chunks before they hit the LLM context window. Claims 60-95% token reduction with comparable answer quality. Ships as a library, a proxy, and an MCP server, so it fits into whatever pipeline you already have.
Explore the radar’s memory
Every pick we’ve ever shipped, plotted as a living constellation. Picks cluster by topic; threads between clusters reveal cross-topic connections. Zoom in to expand a cluster into its picks — drag, filter, click through to the brief.
// Browse
All picks
Archive results
-
chopratejas/headroom
-
chubbyguan/chubbyskills
Pulls content from Douyin, Bilibili, Xiaohongshu, WeChat, X, and podcasts into a personal knowledge base, with auto-routing for image and video and a subtitle-first transcription layer. Surfaces everything through an MCP server so your agent can query it.
-
rahilp/second-brain-cloudflare
A self-hosted memory layer that sits behind any MCP-compatible AI tool — Claude, ChatGPT, Cursor, whatever. Store a fact once, recall it everywhere. Runs on Cloudflare's free tier, so the cost of ownership is zero until you deliberately outgrow it.
-
capture0x/AdStrike
A modular Active Directory red-team framework that wires AI into the full pentest workflow — enumeration, attack-path analysis, Kerberos and ADCS chains, automated reporting, and an MCP server for operator automation. Built for authorized engagements, not script-kiddie runs.
-
tuchg/Lucarne
Bridges your running AI coding sessions — Codex, Claude, and similar — to your phone via Telegram, WeChat, or Feishu. Get notified when the agent needs a decision, approve or reject from your phone, and let it keep running. No hooks, no modified runtimes, no MCP server required.
-
sandiiarov/skill-creator
Converts any MCP server, OpenAPI spec, or GraphQL endpoint into a runnable CLI at runtime — no codegen step, no manual wrapper writing. Useful for any builder who wants to script against an API without maintaining a bespoke client library.
-
Callous-0923/agent-study
A 36-chapter, runnable-Python curriculum covering the full AI agent stack: ReAct loops, MCP and A2A protocols, RAG, DSPy, production observability, and a reverse-engineering walkthrough of Claude Code. Built interview-first, so every chapter ships as a file you can actually execute.
-
jdevalk/specification.website
An open-source website specification covering HTML, accessibility, security, SEO, and agent-readiness — all sourced and MIT-licensed. Useful as a shared checklist for teams who want one authoritative document to argue from instead of a dozen blog posts.
-
butterbase-ai/butterbase-oss
An open-source backend-as-a-service combining Postgres, auth, storage, serverless functions, an AI gateway, and MCP support in one self-hostable stack. Think Supabase but with the AI plumbing already wired in rather than bolted on later.
-
Kaelio/ktx-ai-data-agents-mcp-context-skills
Adds an executable context layer between AI agents like Claude Code or Codex and your data sources, so queries come back accurate instead of hallucinated. Targets analytics and data engineering workflows where MCP alone is not enough to keep the agent on the rails.
-
sansan0/TrendRadar
Self-hostable trend and public-opinion monitor that aggregates multi-platform signals, RSS feeds, and keyword alerts, then pushes AI-summarized briefings to Telegram, Slack, WeChat, and several other channels. Docker-native, data stays on your infra.
-
google-antigravity/antigravity-sdk-python
A Python library for building AI agents on top of Google's Antigravity platform. If you are already evaluating agent frameworks this week, this one is worth a look given the backing and the scope of the API surface it exposes.
-
flipped-aurora/gin-vue-admin
A full-stack admin scaffold built on Gin and Vue 3 with AI-assisted code generation, MCP helpers, JWT auth, RBAC, and a form builder baked in. Primarily aimed at teams building enterprise internal tools in a Chinese-language context. Solid feature list if that stack matches yours.
-
78/xiaozhi-esp32
An MCP-based chatbot firmware for ESP32 microcontrollers. Interesting if you are building embedded voice or chat interfaces on cheap hardware, but niche enough that most SaaS founders will scroll past. Open-source and Chinese-language-primary documentation.
-
PrefectHQ/fastmcp
A decorator-driven Python framework that makes building MCP servers and clients as straightforward as FastAPI made REST, this tool is aimed at AI engineers and indie hackers who want to expose their services as agent-ready tool surfaces without drowning in protocol boilerplate.
-
github/github-mcp-server
Built and maintained by GitHub itself, this MCP server gives AI agents direct, first-party access to repos, issues, pull requests, and Actions runs without any custom glue code. Small SaaS teams curious about autonomous issue triage or AI-assisted PR workflows will want to read the full breakdown before wiring it into production.
-
microsoft/playwright-mcp
An MCP server wrapping Playwright, letting AI agents drive a real browser for scraping, testing, or automation tasks. Useful for any team building agents that need to interact with web UIs that have no API.
-
PDFMathTranslate/PDFMathTranslate
A layout-aware PDF translator built specifically for scientific papers, keeping equations, figures, and column formatting intact across Google, DeepL, Ollama, and OpenAI backends. Researchers and technical PMs who regularly wade through non-English arXiv papers or academic literature will want to see how the Zotero plugin alone could reshape their reading workflow.
-
ChromeDevTools/chrome-devtools-mcp
A coding agent that can truly see and control a live browser is now possible through an MCP server built by the actual Chrome DevTools team, giving AI assistants full access to the real DevTools protocol — DOM inspection, console errors, network requests, and more. Engineers building AI coding assistants or automated QA pipelines should take a close look.
-
punkpeye/awesome-mcp-servers
A curated list of MCP servers for connecting LLMs to external tools and services. Useful as a reference when building agent integrations and you want to know what already exists before rolling your own. Good first stop for AI engineers scoping an MCP-based project.
-
docsagent/docsagent
DocsAgent indexes thousands of local desktop documents and lets you chat with them without sending anything to the cloud. Solid pick for privacy-sensitive teams or solo operators who live in large local document libraries and want a no-leak alternative to uploading files to a hosted service.
-
mukul975/cve-mcp-server
A production-ready server that wires 27 security intelligence tools into Claude at once, pulling from CVE databases, EPSS scoring, Shodan, VirusTotal, MITRE ATT&CK, and more. Security engineers and founders building security-adjacent SaaS will want to see how this collapses a dozen manual lookup tabs into a single conversational workflow.
-
nduckmink/arkon
A self-hosted knowledge hub that centralizes your team's RAG contexts, access policies, and AI skills in one place, then connects everything to Claude and other LLMs through a single Model Context Protocol integration. If your team is tired of every developer building their own ad-hoc AI pipelines, this is worth a closer look.
-
irinabuht12-oss/google-meta-ads-ga4-mcp
A single server connection that pipes Google Ads, Meta Ads, and GA4 data directly into Claude, Cursor, or n8n, giving founders running paid acquisition over 250 tools to query performance, adjust budgets, and trace funnels without switching between dashboards — compelling for small teams already deep in an AI-native workflow.
-
ThinkWatchProject/ThinkWatch
A self-hosted gateway that puts one authenticated layer in front of every LLM API your team touches, with per-user rate limits, audit logs, and cost tracking broken down by team or feature. SaaS teams tired of scattered API keys and surprise billing spikes will want to see how this compares to rolling their own solution.