Trending AI agent projects this week

Data as of 2026-06-13 · first weekly scoring run pending · updated weekly

Early-launch ranking: ordered by real stars gained over our first 4 days of daily tracking. The full acceleration model described below activates automatically as the dataset matures — nothing here is ranked by total popularity.

A project is trending when its star growth is accelerating — gaining at least twice as fast as the previous week — or when it is in a sustained large spike of 500+ stars in a week, with a composite growth score of 70 or higher. This is acceleration, not size: a small project doubling its pace can out-rank a giant.

MCP servers: 11Agent frameworks: 3Agent SDKs: 1AI agents: 5
#ProjectCategoryΔ★ 7dTotal ★
1microsoft/markitdown mcp-server+2,435152,233
2Firecrawl ai-agent+1,161132,068
3Scrapling MCP Server mcp-server+72663,341
4Claude Code agent-sdk+649132,109
5Browser Use ai-agent+48998,544
6io.github.ruvnet/claude-flow mcp-server+42159,197
7io.github.IncorporatedPartners/labelhead-artist-momentum mcp-server+38770,231
8io.github.t8y2/dbx mcp-server+3845,155
9OpenHands ai-agent+33376,700
10LiteLLM agent-framework+32350,211
11Codebase Memory mcp-server+2953,404
12LangGraph agent-framework+24334,583
13Chrome DevTools MCP mcp-server+23943,495
14LangChain agent-framework+217139,163
15io.github.butterbase-ai/mcp mcp-server+2112,019
16crewAI ai-agent+19753,392
17Agent Skills Search Server mcp-server+16020,407
18Context7 mcp-server+15357,261
19cal.com ai-agent+15245,477
20MCP Reference Servers mcp-server+14987,147

Methodology

Acceleration ratio
Stars gained this week divided by stars gained over the previous two weeks. A ratio of 2.0 means growth doubled. Tiny absolute numbers are gated out so noise never trends.
Sustained-spike path
A project gaining 500+ stars in a week stays trending even after acceleration flattens — a peak is still a peak.
Growth score threshold
The composite growth score (acceleration 40%, relative velocity 30%, downloads 20%, repo activity 10%) must be 70 or higher.
All signals are computed from daily snapshots of stars, forks, releases, and npm/PyPI downloads, aggregated weekly. Every number above is traceable to a snapshot date — nothing is editorial, and rankings cannot be bought. Full details on the methodology page.