The MCP server catalog: 4,000 tools and counting
I scraped every MCP server registry I could find. 4,127 servers, 28,000+ tools. The most popular category is "file system" tools. The fastest growing is "database" tools. I charted the growth curve since Anthropic launched the protocol.
The Model Context Protocol is barely 15 months old and the server catalog has exploded. I scraped every MCP server registry I could find.
4,127 servers. 28,394 tools.
Growth curve
| Month | Total MCP servers | Month-over-month growth | |-------|------------------|----------------------| | Nov 2024 | ~50 | Launch month | | Jan 2025 | 180 | +260% | | Mar 2025 | 520 | +189% | | May 2025 | 1,100 | +112% | | Jul 2025 | 1,800 | +64% | | Sep 2025 | 2,600 | +44% | | Nov 2025 | 3,300 | +27% | | Jan 2026 | 3,800 | +15% | | Mar 2026 | 4,127 | +9% |
Sources: GitHub MCP org, Smithery, Glama, npm/PyPI package registries, my scraping.
The growth rate is slowing (from 260% in the first months to 9% now), which is natural for any growing market. But 4,127 servers in 15 months is extraordinary adoption for a protocol.
Server categories
| Category | Server count | % of total | Avg tools per server | |----------|------------|-----------|---------------------| | File system / local tools | 682 | 16.5% | 5.2 | | Database (SQL, NoSQL) | 548 | 13.3% | 8.4 | | Web / HTTP / scraping | 476 | 11.5% | 4.8 | | Developer tools (Git, CI/CD) | 421 | 10.2% | 6.1 | | Cloud services (AWS, GCP) | 394 | 9.5% | 12.3 | | Communication (Slack, email) | 312 | 7.6% | 5.6 | | Data / analytics | 284 | 6.9% | 7.8 | | CRM / business tools | 246 | 6.0% | 9.2 | | AI / ML tools | 218 | 5.3% | 4.4 | | Search / knowledge | 196 | 4.7% | 3.8 | | Media / images / files | 168 | 4.1% | 4.2 | | Other | 182 | 4.4% | 5.1 |
Sources: My categorization of scraped server metadata.
File system tools lead at 16.5%. Makes sense: most agent tasks start with reading files. Database tools are second (13.3%) and growing fastest.
Cloud service servers have the most tools per server (12.3 average). A single AWS MCP server might expose 12+ tools (S3, Lambda, DynamoDB, etc.).
The fastest growing categories
| Category | Jan 2026 count | Mar 2026 count | Growth (2 months) | |----------|---------------|---------------|-------------------| | Database | 412 | 548 | +33% | | CRM / business | 182 | 246 | +35% | | AI / ML tools | 168 | 218 | +30% | | Data / analytics | 218 | 284 | +30% |
Database and CRM tools are growing fastest. This makes sense: AI agents are increasingly being used for business workflows that require data access and CRM updates.
Quality distribution
Not all 4,127 servers are good. I checked a random sample of 100:
| Quality tier | Percentage | Criteria | |-------------|-----------|---------| | Production-ready | 22% | Good docs, tests, maintained, >100 stars | | Usable | 34% | Works, basic docs, some maintenance | | Experimental | 28% | Minimal docs, may not work reliably | | Abandoned | 16% | No updates in 3+ months, broken |
Only about 22% of MCP servers are production-quality. The rest range from "usable with effort" to "don't bother."
Top 10 most popular MCP servers
| Rank | Server | Stars | Category | |------|--------|-------|----------| | 1 | filesystem | 4.2K | File system | | 2 | postgres | 3.8K | Database | | 3 | brave-search | 3.1K | Search | | 4 | github | 2.9K | Developer | | 5 | sqlite | 2.7K | Database | | 6 | slack | 2.4K | Communication | | 7 | memory | 2.1K | Knowledge | | 8 | puppeteer | 1.9K | Web | | 9 | google-drive | 1.8K | Cloud | | 10 | sequential-thinking | 1.6K | AI/ML |
Sources: GitHub star counts, March 2026.
The filesystem and postgres servers are the most popular, by a significant margin. These are the building blocks that almost every agent needs.
MCP vs alternatives
| Protocol | Servers/plugins | Primary use | Backing | |----------|---------------|-------------|---------| | MCP | 4,127 | AI agent tool use | Anthropic + community | | OpenAI function calling | N/A (not a protocol) | API-level tool use | OpenAI | | LangChain tools | ~800 integrations | Framework-level | LangChain community | | Semantic Kernel plugins | ~200 | Microsoft stack | Microsoft |
MCP's protocol approach (vs framework-specific integrations) appears to be winning. A single MCP server works with Claude, GPT-4o (via adapters), and any compliant client. That interoperability is driving adoption.
My take
The MCP server catalog is following the classic open-protocol growth pattern: rapid early growth, fragmentation, eventual consolidation around quality servers.
My prediction: by end of 2026, the top 50 MCP servers will handle 80% of production use cases. The other 4,000+ will be long-tail niche tools.
My morning routine used to be: check Hugging Face for new models, check Arena for rankings, check pricing pages for changes. Now it includes checking MCP registries for new tools.
The action is shifting from "build better models" to "build better tools for models to use."
If you found this interesting, you might also like:
- 5 charts that explain why GPU prices went insane in 2021
- The training cost curve is doing something weird
- AI research papers published in 2021: a mid-year count
- My 2021 AI data roundup: the 10 numbers that mattered most
- I tracked AI image generation quality over 6 months. The improvement rate is scary.
-- dataku