Five agent repos gaining momentum. Every one measured.

Ranked by stars gained over the last 7 days, growth rate and recent commit activity.

Short candidate pool today: 3 of 5 picks were filled with relaxed rules (3 by star gains since the last snapshot).

Rank 1

google-gemini/gemini-cli

Gemini CLI is an open-source terminal agent that gives direct command-line access to Gemini models, supporting codebase querying and editing, non-interactive scripting, and MCP servers for added capabilities. It authenticates via Google OAuth, a Gemini API key, or Vertex AI, and ships through npm, Homebrew, and MacPorts with weekly preview/stable and daily nightly release channels.

Developers building terminal-based or scriptable agents can look at its MCP integration and non-interactive mode, but note the README says it has been replaced by Antigravity CLI for unpaid and Google One users and is not optimized for the latest Gemini models.

★ 107,532
+★ 122 in 7 days
0.1% growth
Apache-2.0
README ≈ 4,945 tokens to load into context

AI-written summary

Score80.0

Rank 2

Significant-Gravitas/AutoGPT

AutoGPT is a platform for creating and coordinating role-based AI 'Experts' (such as content, sales, and executive assistant specialists) that you brief in plain English, with a coordinator called Otto handing work between them. It is offered as a paid hosted service or can be self-hosted from a single Docker image (significantgravitas/autogpt), where you supply your own model credentials.

Developers who want to build or run multi-agent workflows on their own infrastructure, or study a platform with configurable Experts, connected apps, and spending controls, may find it relevant.

★ 187,713
+★ 102 in 7 days
0.1% growth
No licence declared
README ≈ 3,940 tokens to load into context

AI-written summary

Score20.0

Rank 3

mksglu/context-mode

Context Mode is an MCP server (with hooks) that keeps raw tool output out of an AI coding agent's context window, tracks session events in SQLite/FTS5 so relevant state can be retrieved after conversation compaction, and pushes the agent to write scripts that return only results. The README cites a 315 KB to 5.4 KB reduction and says it supports 17 client platforms.

Developers building or running coding agents that burn context on large MCP outputs (Playwright snapshots, GitHub issues, logs) or lose track of in-progress work after compaction should look at it.

★ 31,859
No licence declared

AI-written summary

Score58.6

Rank 4

mvanhorn/last30days-skill

last30days is an AI agent skill that searches Reddit, X, YouTube, TikTok, Hacker News, Polymarket, GitHub, and other sources in parallel for a given topic. It scores results by engagement signals such as upvotes, likes, and market odds, then synthesizes them into a single brief covering the last 30 days.

Developers building agents that need recent, community-sourced research across several platforms can install it as a skill for Claude Code, Codex, Cursor, Gemini CLI, and other Agent Skills hosts.

★ 65,148
MIT
README ≈ 12,233 tokens to load into context

AI-written summary

Score51.7

Rank 5

DeusData/codebase-memory-mcp

codebase-memory-mcp is an MCP server written in C that builds a persistent local knowledge graph connecting code, documentation, configuration, and infrastructure, exposed through 17 MCP tools and a CLI. It also ships CBM Lens, an optional open-beta (v0.0.1) visual companion, plus an installer for 45 agent client integrations and change-aware test selection.

Developers building or using coding agents who want structural repository context (call tracing, change-impact analysis) from a local index, with no hosted service or API key, should look at it.

★ 47,269
MIT

AI-written summary

Score50.7