Growth trajectory

The MCP ecosystem grew at extraordinary speed in early 2025 and is now decelerating toward a normal compounding rate. Q1 2025 saw a 1,600% jump — that's what protocol announcement effect looks like. Q1 2026 added 4,532 new repositories, a 21% quarter-over-quarter increase. That's healthy, sustainable growth for an ecosystem this size.

The deceleration from 188% (Q2 2025) to 21% (Q1 2026) is not a sign of stagnation. It's what ecosystem maturation looks like. The initial wave was developers trying the protocol. Q1 2026 growth is production adoption — companies and maintainers building tools they intend to ship.

If current trends hold, the index will cross 30,000 repositories by Q2 2026. The skills registry count (3,124) is growing faster proportionally — Glama and skills.sh added significantly more listings in Q1 2026 than in any prior quarter.

Top 10 by AgentRank score

AgentRank scores 2,000 MCP tools daily using five signals: stars (15%), freshness (25%), issue health (25%), contributors (10%), and inbound dependents (25%). As of March 19, 2026, 34 tools score 90 or above — the Elite tier. Here are the top 10.

The top 10 spans game-dev integration (Unity MCP tools at #1 and #5), enterprise SaaS connectivity (Microsoft, Laravel), frameworks (fastmcp, mcp-go), and browser automation (Playwright). No single category dominates — breadth is the pattern.

microsoft/playwright-mcp (#9) has the most GitHub stars of any tool in the index at 29,150 — yet ranks 9th, not 1st. This illustrates the score's intent: stars alone don't win. playwright-mcp scores high on freshness and issue health too, but several newer tools with lower star counts have even better combined signals.

Category breakdown

The AgentRank index covers 1,735 items with category labels (110 tools + 1,625 skills as of March 19, 2026). AI tools and MCP servers account for over 80% of the scored index. Code assistants are the third-largest category — this is where star counts are highest and install velocity is fastest.

The AI Tool category at 47% is the dominant bucket. This covers general-purpose LLM integrations, RAG pipelines, chat bridges, and prompt utilities that don't fit a narrower domain. As the index matures, these will migrate into more specific categories.

Code Assistant at 14% punches above its weight in install volume — this is where the skills.sh and Glama install counts concentrate. A single code-assistant skill averages roughly 10x the installs of a general AI tool.

Browse by category: Database · DevOps · AI/ML · Browser · Security

For ranked picks in each category, see our best-of lists: Database · DevOps · AI/ML · Browser Automation · Security · Code Generation

Biggest movers this quarter

These are the five tools with the largest star growth from Q4 2025 to Q1 2026 mid-point. Star velocity is a leading indicator of score improvement — it usually precedes the issue health and dependency signals that drive AgentRank scores higher.

CoplayDev/unity-mcp is the standout story of Q1. A community-built tool with no corporate backing that holds the #1 AgentRank score (98.7) and grew 126% in stars this quarter. It reached the Unity game-dev community organically and spread from there. This is what the top of the index looks like when community momentum beats corporate investment.

oraios/serena is the fastest-growing tool in the index at +130% — from roughly 9,400 to 21,673 stars in one quarter. A semantic code editing agent, it's positioned directly at the "AI pair programmer that understands your whole codebase" use case that the market is hungry for.

Notable new entrants

Three new tools entered the top-10 scoring tier this quarter, all with strong institutional backing. These were either launched or substantially v2'd during Q1 2026.

laravel/boost (#3 overall, score 96.9) is the most surprising entrant. PHP was never an early adopter of AI tooling, but the Laravel Foundation moved fast once the spec was stable. Official maintainer backing gives it exceptional issue health and freshness signals. It's a reminder that ecosystem coverage is not determined by language popularity alone.

microsoft/azure-devops-mcp (#2 overall, score 97.3) launched in Q1 2026 and immediately showed what corporate release quality looks like: 1,400+ stars, a responsive issue tracker, and a team large enough to ship point releases consistently.

homeassistant-ai/ha-mcp (score 94.3) is the standout community entry — an official Home Assistant MCP server that connects AI agents to smart home state. Consumer IoT as an MCP target was not on anyone's 2025 roadmap, and it's now top-10 by score.

Language & license trends

Python's share of the scored index is declining — not because Python projects are disappearing, but because TypeScript and Rust are growing faster. This reflects a maturation in who's building MCP tools: fewer ML researchers experimenting, more product engineers shipping production tooling on edge runtimes. Data from the AgentRank tools table (110 scored tools, March 2026).

TypeScript is the biggest gainer (+5.1pp to 30.9%), narrowing the gap with Python significantly. The Vercel/Cloudflare/edge runtime ecosystem is TypeScript-native, and the agent-tooling layer is increasingly deployed on edge workers, not local Python scripts. This shift was slower than expected in 2025 — it's accelerating now.

Rust is up to 5.5%, more than doubling its share. Rust MCP tools are disproportionately represented in the Elite tier (90+ score) — the language attracts maintainers who care enough to do it properly. C# reached 3.6% for the first time, driven by enterprise .NET shops building MCP tooling for Azure and Microsoft 365.

On licensing: MIT dominates at 60.9% (67 of 110 scored tools). 28.2% have no license — that's a risk flag the AgentRank score penalizes implicitly through contributor and issue health signals. Apache-2.0 covers the rest (5.5%). Open-source by default remains the norm in MCP tooling.

Freshness snapshot

Freshness accounts for 25% of the AgentRank score — the highest weight of any single signal. The 2,000 ranked MCP tools span a wide range of activity levels. Score 90+ tools are essentially all actively maintained; the long tail shows the typical open-source decay pattern.

The skill install data is the most concrete freshness signal in the index. skills.sh tracks 12.4 million installs across 1,125 skills; Glama tracks 2.5 million across 500 MCP tools. These are not vanity metrics — install counts directly reflect tool usage at runtime, not just passive GitHub stars.

The score distribution tells the true story of ecosystem health: 82% of ranked tools cluster in the 50–69 "Active" band — above a baseline threshold but not yet Elite. The gap between a 65-scoring tool and a 95-scoring tool is almost always freshness and issue health, not stars.

Q2 2026 predictions

1. A2A crosses 5% of the index

A2A (agent-to-agent) tools currently represent 3.2% of indexed repositories. The spec is maturing fast and Google's A2A launch created a second wave of ecosystem tooling. By the end of Q2 2026, we expect A2A tools to cross 5% of the index — roughly 1,500+ repositories. The benchmark: MCP crossed 5% of "AI tooling" GitHub search results in its third month post-launch. A2A is on a similar curve. See our A2A vs MCP protocol comparison for a full breakdown of how the two protocols differ.

2. TypeScript overtakes Python as the dominant language

At Q1 2026 growth rates, TypeScript reaches Python's current share (~38%) by Q3 2026. The inflection point may come sooner. Vercel, Cloudflare, and the VS Code/Cursor ecosystem are all TypeScript-native, and the agent-tooling layer is increasingly built for deployment on edge runtimes, not local Python scripts.

3. Enterprise security category doubles

The Security & Compliance category is growing faster than any other vertical by repository count. Enterprise security teams are building MCP servers for SAST integration, secrets scanning, compliance audit, and cloud posture management. Expect the current 320 repos to approach 700 by end of Q2 2026. The SOC 2 / ISO 27001 compliance layer is coming to agent tooling.

4. Installation counts diverge further from star counts

The top 5 tools by Glama installs already account for more than 60% of total install volume. As MCP moves into production, installation concentration will increase — users default to known-good tools rather than experimenting. This means the mid-tier tools (score 40–60) will need a clear differentiation narrative to avoid being bypassed entirely.

5. Registry consolidation

There are currently 6+ MCP registries (Glama, skills.sh, Smithery, the official registry, npm tags, and aggregators like AgentRank). By Q3 2026, expect clear winners to emerge — likely 2–3 dominant registries with cross-listing support. The winner will be the one that AI coding tools use by default.