New on AgentIndex: 13 Tools Worth a Look (September–October 2026)

Between September 7 and October 7, 2026, AgentIndex added 299 open-source tools to the directory. That is too many to browse one by one, so this roundup picks 13 that stood out for usefulness, momentum or a fresh approach, grouped by what they help you do.

A note on method: "new" here means newly listed on AgentIndex. Most of these projects were created within the last year, but a few are older projects that recently added MCP or agent features, and we say so where that applies. Star counts and licenses come from GitHub on October 7, 2026, and descriptions are based on each project's own README. We skipped tools that were off-topic for AI agents and duplicates of existing listings.

Where the new tools came from

By the directory's automatically assigned categories, the largest groups among the 299 additions were Workflow Automation (60), Dev Tooling (47), API Integration (36), RAG / Knowledge Base (27), Multi-Agent (27), MCP Servers (23) and Memory & Context (21). Two trends are visible: many developer tools now ship an MCP server as a standard feature, and a growing number of projects focus specifically on giving coding agents better memory and context.

Memory and context for coding agents

aoci-code (1.2k ★, created August 2026). A local-first MCP server and CLI written in Go that maintains a persistent, Git-versioned map of your codebase and database schema. The idea is that Claude Code, Codex, Cursor or opencode read this map before touching anything, instead of rediscovering the project on every task. Check the license terms before adopting it; GitHub does not identify a standard license.

Hippo (773 ★, created March 2026, MIT). Long-term memory for coding agents that works across tools such as Cursor, Claude Code and Codex. It describes a brain-inspired design with decay, retrieval strengthening and consolidation, and keeps data local.

Gentle-AI (7.6k ★, created February 2026, MIT). Rather than being another agent, it configures the ones you already use (Claude Code, Cursor, OpenCode, Codex, Pi and more) with persistent memory, curated skills, MCP servers and personas, without locking you into one agent.

Ref Tools MCP (1.2k ★, MIT). Gives coding agents accurate documentation for public and private libraries while trying not to waste the context window. A good alternative or complement to Context7.

Turning data into agent-ready context

PipesHub (3.8k ★, Apache-2.0). A self-hosted "context layer" that turns company knowledge from Slack, Google Drive, Jira, GitHub, Microsoft 365 and 40+ other connectors into a permission-aware workspace that agents can search, navigate and cite. It ships MCP support, SDKs and built-in agents.

anymd (1.0k ★, MIT). A fast Rust MCP server and CLI that converts almost any file into clean Markdown for agents: PDF, Word, PowerPoint, Excel, EPUB, HTML and web pages, images via OCR, and audio or video metadata and transcripts. It runs locally with no API key.

crw (1.1k ★, created March 2026, AGPL-3.0). A lightweight web scraper, crawler and search API in Rust with an MCP server and a Firecrawl-compatible API. The project reports faster results than Firecrawl and Tavily in its own benchmarks; treat those numbers as the authors' claims and test on your own workload. Note the AGPL license if you plan to modify and offer it as a service.

korean-law-mcp (2.6k ★, MIT). An MCP server for searching Korean statutes, precedents and ordinances from the national legal information service, including citation verification. A good example of a focused, domain-specific server.

Automation beyond the browser

docker-android (15.9k ★). A long-running project (since 2016) that runs Android emulators in Docker with noVNC access and video recording, and now includes an MCP server and AI-agent support. Useful for letting agents test mobile apps in a reproducible environment.

A9T9 RPA (2.1k ★). Another established project (since 2017) for browser and desktop automation, now with an MCP server for AI assistants. You can record macros or write JavaScript, and it uses OCR and image recognition to control desktop apps.

Pascal Editor (24.7k ★, MIT). An open-source 3D architectural editor with a local CLI and MCP tools, designed for workflows shared between people and AI agents. The most-starred addition of the month.

Security and local AI

AiSOC (2.4k ★, created May 2026, MIT). An open-source AI security operations center: alert fusion, LLM-agent triage, MITRE ATT&CK-based investigation and a replayable record of every agent decision. It is self-hostable, can run without API keys and ships an MCP server for Claude, Cursor and Continue.

Off Grid AI (OGAM) (3.2k ★, created January 2026, MIT). A private, offline AI toolkit for phones and Macs that combines chat with GGUF models, vision, Whisper speech-to-text, Stable Diffusion image generation and tool calling. No account or API key, and data stays on the device.

What to take away

• MCP is becoming a default feature. Older automation projects such as docker-android and A9T9 RPA are adding MCP servers, which makes them usable from any MCP-compatible agent.
• Context is the new bottleneck. Several of the most interesting additions (aoci-code, Hippo, Gentle-AI, Ref Tools, PipesHub) exist to give agents better memory and more relevant context rather than more raw capability.
• Check licenses on new projects. Some newer tools use AGPL or have no standard license detected. That is fine for experiments, but read the terms before building on them.

Browse the newest additions any time, or read our guide on choosing an MCP server before connecting any of these to your own tools.