State of AI Assets — Q1 2026
The first open census of the AI agent ecosystem. 5M+ assets. 143K agents & tools. All trust scored.
We built Nerq to answer a simple question: what's actually out there in the AI agent ecosystem, and how much of it can you trust?
After crawling six registries — GitHub, npm, PyPI, HuggingFace, Docker Hub, and MCP registries — we now have an answer. Here's what we found.
The numbers
4,919,620 AI assets indexed. Of those, 143,642 are the executable components — agents, tools, and MCP servers — that power the emerging agentic economy.
Every asset receives a Trust Score (0–100) based on security, maintenance, popularity, documentation, and ecosystem signals.
What's in the index
Most of the 4.9M assets are ML models (2.6M), HuggingFace Spaces (1M), and datasets (795K). The actionable layer — what actually does things — is smaller but growing fast:
What agents do
Coding dominates. Infrastructure and DevOps are second and third — agents are increasingly used for operational automation, not just chat.
How they're built
Anthropic SDK leads. OpenAI is close behind. LangChain is the dominant orchestration framework. MCP (Model Context Protocol) already ranks 4th — a strong adoption signal for Anthropic's tool protocol.
Python dominates at 51% of agents with known languages. TypeScript is second at 22%, driven by MCP server development and npm packages.
Where they come from
Trust & quality
12.9% of agents score 70+ (high trust). 87.1% score 40–69 (medium). Almost none score below 40.
The average trust score is 65.5 / 100 — decent but with significant room for improvement across the ecosystem.
Trust is computed from five pillars:
Top 10 most trusted agents
Top 10 MCP servers
Try it
All data is free, no auth required.
Search agents:
curl https://nerq.ai/v1/agent/search?q=la...n&min_trust=60
Check trust on any agent:
curl https://nerq.ai/v1/agent/kya/SWE-agent
Get ecosystem stats:
curl https://nerq.ai/v1/agent/stats
Benchmark a category:
curl https://nerq.ai/v1/agent/benchmark/coding
Links
Data as of 2026-03-09. Nerq indexes continuously — numbers update daily.
More...
The first open census of the AI agent ecosystem. 5M+ assets. 143K agents & tools. All trust scored.
We built Nerq to answer a simple question: what's actually out there in the AI agent ecosystem, and how much of it can you trust?
After crawling six registries — GitHub, npm, PyPI, HuggingFace, Docker Hub, and MCP registries — we now have an answer. Here's what we found.
The numbers
4,919,620 AI assets indexed. Of those, 143,642 are the executable components — agents, tools, and MCP servers — that power the emerging agentic economy.
Every asset receives a Trust Score (0–100) based on security, maintenance, popularity, documentation, and ecosystem signals.
| Total AI assets | 4,919,620 |
| Agents & tools | 143,642 |
| MCP servers | 17,468 |
| Avg trust score | 65.5 / 100 |
| High trust (70+) | 12.9% |
What's in the index
Most of the 4.9M assets are ML models (2.6M), HuggingFace Spaces (1M), and datasets (795K). The actionable layer — what actually does things — is smaller but growing fast:
- 66,092 agents (autonomous software that acts)
- 60,082 tools (packages agents use)
- 17,468 MCP servers (tool servers for LLMs)
What agents do
Coding dominates. Infrastructure and DevOps are second and third — agents are increasingly used for operational automation, not just chat.
| Coding | 10,939 |
| Infrastructure | 3,552 |
| DevOps | 3,524 |
| Communication | 2,955 |
| AI tool | 2,262 |
| Finance | 2,132 |
| Research | 1,909 |
| Content | 1,277 |
| Marketing | 1,253 |
| Data | 1,199 |
| Security | 1,160 |
| Healthcare | 791 |
| Education | 751 |
| Productivity | 718 |
How they're built
Anthropic SDK leads. OpenAI is close behind. LangChain is the dominant orchestration framework. MCP (Model Context Protocol) already ranks 4th — a strong adoption signal for Anthropic's tool protocol.
| Anthropic | 7,072 |
| OpenAI | 5,927 |
| LangChain | 2,546 |
| MCP | 1,932 |
| Ollama | 1,785 |
| HuggingFace | 1,126 |
| AutoGen | 1,065 |
| CrewAI | 780 |
| LlamaIndex | 424 |
Python dominates at 51% of agents with known languages. TypeScript is second at 22%, driven by MCP server development and npm packages.
Where they come from
| HuggingFace | 35.5% |
| GitHub | 21.4% |
| npm | 21.4% |
| PyPI | 17.3% |
| MCP registries | 1.4% |
Trust & quality
12.9% of agents score 70+ (high trust). 87.1% score 40–69 (medium). Almost none score below 40.
The average trust score is 65.5 / 100 — decent but with significant room for improvement across the ecosystem.
Trust is computed from five pillars:
- Security (30%) — vulnerabilities, dependency audit, code patterns
- Maintenance (25%) — commit recency, release frequency, issue response
- Popularity (20%) — stars, downloads, forks, community size
- Documentation (15%) — README quality, API docs, examples
- Ecosystem (10%) — protocol support, integrations, interoperability
Top 10 most trusted agents
| SWE-agent/SWE-agent | 92.5 | 18,516 |
| microsoft/qlib | 92.4 | 37,615 |
| nanoclaw | 92.1 | 7,735 |
| FunnyWolf/agentic-soc-platform | 91.3 | 579 |
| ccmanager | 90.9 | 831 |
| microsoft/azure-devops-mcp | 90.3 | 1,291 |
| opal | 90.2 | 5,422 |
| GoogleCloudPlatform/agent-starter-pack | 90.1 | 5,761 |
| QwenLM/qwen-code | 89.7 | 20,060 |
| RooCodeInc/Roo-Code | 89.5 | 22,330 |
Top 10 MCP servers
| williamzujkowski/strudel-mcp-server | 92.9 | 158 |
| laravel/boost | 91.2 | 3,275 |
| harbor | 90.5 | 2,424 |
| laravel/mcp | 90.0 | 679 |
| CursorTouch/Windows-MCP | 89.0 | 4,390 |
| Ansvar-Systems/EU_compliance_MCP | 88.8 | 45 |
| 54yyyu/zotero-mcp | 88.1 | 1,461 |
| tavily-ai/tavily-mcp | 88.1 | 1,218 |
| cyproxio/mcp-for-security | 88.1 | 553 |
| minecraft-mcp-server | 87.8 | 467 |
Try it
All data is free, no auth required.
Search agents:
curl https://nerq.ai/v1/agent/search?q=la...n&min_trust=60
Check trust on any agent:
curl https://nerq.ai/v1/agent/kya/SWE-agent
Get ecosystem stats:
curl https://nerq.ai/v1/agent/stats
Benchmark a category:
curl https://nerq.ai/v1/agent/benchmark/coding
Links
- nerq.ai — search the index
- API docs
- KYA — Know Your Agent — due diligence reports
- Live stats
- Full report with all 20 top agents
- MCP server on Smithery
Data as of 2026-03-09. Nerq indexes continuously — numbers update daily.
More...