This Week in MCP
Weekly data from the AgentRank index. Auto-generated from real GitHub signals every week. 25,638 tools indexed — 263 new entries this week, and microsoft/playwright-mcp just entered the top 10 with 29K stars.
Ecosystem stats
Only 149 tools score above 80 — the top 0.6% of the index. An additional 352 tools score 60–80. The long tail is real: 18,870 tools (73.6%) score below 40, reflecting the reality that most MCP repos are experimental or inactive.
New tools this week
263 new repositories entered the index this week. Here's how they break down by language:
| Category | Count | Highlights |
|---|---|---|
| AI agent infrastructure | ~40 | Memory systems, agent frameworks, security scanners |
| Media & creative tools | ~35 | ComfyUI bridges, image generation, video workflows |
| Dev tooling | ~80 | CLI wrappers, TUI apps, debug tools with MCP servers |
| Platform integrations | ~60 | Service wrappers, API bridges, SaaS connectors |
| Other / experimental | ~48 | Research projects, forks, proof-of-concepts |
Notable newcomers
Five new entries worth watching from this week's batch:
Security scanner for AI agents, MCP servers, and plugins. 30 rules, AST taint tracking, cross-file analysis, and kill chain detection. Billed as a free alternative to Snyk Agent Scan. Security tooling for the AI agent layer is a new category — watch for more here.
AI Agent Fleet Memory System combining Qdrant, Mem0, and Neo4j/Graphiti. Composite scoring, compaction engine, temporal knowledge graph, multi-claw federation, and sleep-time reflection — all exposed as an MCP server. Most complex memory architecture we've indexed to date.
Bridges NotebookLM's research capabilities into a Claude Code skill + MCP server pipeline: NotebookLM researches, Claude synthesizes and publishes. A workflow-chain pattern that will likely be replicated across other tool pairs.
Converts ComfyUI image/video generation workflows into parameter-configurable MCP tools for AI agents. Released in both TypeScript and Python simultaneously. ComfyUI has 60K+ stars — if this gains traction it opens a large new use case for agent-driven media generation.
Production-ready AI agent framework with semantic memory, multi-agent mesh, MCP server, intelligent routing, governance, and 67+ platform integrations. A full-stack agent framework with MCP as a first-class citizen — rather than MCP bolted on top.
Biggest gainers
Tools with the largest score increase compared to the previous week's snapshot.
| Repository | Score change | Prev rank | Now |
|---|---|---|---|
| CoderGamester/mcp-unity | +32.79 | #43 | #4 |
| homeassistant-ai/ha-mcp | +32.07 | #72 | #16 |
| microsoft/azure-devops-mcp | +32 | #32 | #2 |
| samanhappy/mcphub | +31.91 | #50 | #9 |
| jgravelle/jcodemunch-mcp | +31.81 | #68 | #18 |
The standout story: microsoft/azure-devops-mcp jumped 30 spots to reach #2 — the highest any enterprise-backed tool has ranked in this index. It now sits between two community-built Unity tools (CoplayDev at #1, CoderGamester at #4), which is an unusual composition for the top 5.
Current top 10
The highest-scoring tools in the index, scored by stars, freshness, issue health, contributors, and dependents.
| # | Repository | Score | Stars | Language |
|---|---|---|---|---|
| #1 | CoplayDev/unity-mcp | 98.67 | 7,003 | C# |
| #2 | microsoft/azure-devops-mcp | 97.17 | 1,406 | TypeScript |
| #3 | laravel/boost | 96.94 | 3,333 | PHP |
| #4 | CoderGamester/mcp-unity | 96.26 | 1,412 | C# |
| #5 | mark3labs/mcp-go | 96.24 | 8,353 | Go |
| #6 | Pimzino/spec-workflow-mcp | 95.89 | 3,999 | TypeScript |
| #7 | PrefectHQ/fastmcp | 94.95 | 23,775 | Python |
| #8 | zcaceres/markdownify-mcp | 94.60 | 2,449 | TypeScript |
| #9 | samanhappy/mcphub | 94.37 | 1,893 | TypeScript |
| #10 | microsoft/playwright-mcp | 94.23 | 29,150 | TypeScript |
Trends to watch
1. The Unity MCP ecosystem is real
Two Unity-focused MCP servers now sit in the top 4: CoplayDev/unity-mcp (#1) and CoderGamester/mcp-unity (#4). Both score above 96, driven by strong freshness, active issue management, and growing contributor counts. Game engine integration is an underrated use case for MCP — Unity's scripting model maps cleanly to tool-call patterns, and agent-controlled game objects are compelling for both dev tooling and automated testing.
2. Enterprise tooling entering the top tier
microsoft/azure-devops-mcp reached #2 this week with a 32-point score jump. Enterprise tools historically lag in the index because they score lower on the "dependents" signal (fewer repos depend on them vs. SDK-style tools). This jump suggests the freshness and issue health signals are carrying it — likely reflecting active development at Microsoft. If Azure DevOps MCP goes GA, it could stay in the top 5 on volume of usage signals alone.
3. Most of the index is stale
65.3% of indexed repositories (16,746 out of 25,638) have no commit activity in the last 60 days. This is the silent majority of the MCP ecosystem: experimental repos, tutorials, weekend projects, and abandoned forks. The AgentRank scoring model penalizes this heavily via the freshness signal (25% of total score). Only 150 tools currently score above 80.
4. Microsoft now holds two spots in the top 10
microsoft/playwright-mcp entered the top 10 this week at #10 (94.23 score, 29,150 stars) — joining microsoft/azure-devops-mcp at #2. With 29K stars, playwright-mcp is the most-starred repo in the top 10 by a wide margin. It's a web automation server that gives AI agents full Playwright browser control through MCP. The score reflects its freshness, issue activity, and enormous dependent footprint. Two Microsoft tools in the top 10 is a new development for this index.
5. AI agent infrastructure is emerging as a category
This week's new entries included a security scanner for agents (agent-shield), a fleet memory system (memclawz), a production agent framework (mayros), and ComfyUI workflow bridges. These are tools that operate one layer above standard MCP servers — they use MCP as transport but solve higher-level agent coordination problems. Watch for this category to grow fast in Q2.
Language breakdown (top 5)
| Language | Tools | Share | Bar |
|---|---|---|---|
| Python | 9,876 | 45.1% | |
| TypeScript | 7,008 | 32.0% | |
| JavaScript | 3,097 | 14.1% | |
| Go | 1,227 | 5.6% | |
| Rust | 669 | 3.1% |
Python and TypeScript together account for 77% of the index. The split reflects the ecosystem's two primary audiences: Python-first data scientists and ML engineers, and TypeScript-first web/fullstack developers. Go is punching above its weight in the top rankings — mark3labs/mcp-go (#5) leads the category.
Tweet this week
Auto-generated tweet from this week's top insight:
This week in MCP: 263 new tools indexed (25,638 total). microsoft/playwright-mcp entered the top 10 with 29,150 stars — the most-starred repo in the top 10 by a wide margin. Two Microsoft tools now in top 10. Best newcomers: agent-shield (security scanner for AI agents) and memclawz (fleet memory system with temporal knowledge graph). Full report: agentrank-ai.com/blog/this-week-in-mcp-2026-03-18/
Get the weekly AgentRank digest
Top movers, new tools, ecosystem insights — straight to your inbox.