Your AI agent picks an MCP server, calls its tools, and acts on the results. But should it?
We built Dominion Observatory to answer that question with data. After indexing every public MCP server across Smithery, mcp.so, Glama, the official registry, and GitHub, we now track 14,824 servers with behavioral trust scores.
Here's what the data says as of May 2026.
The Numbers
| Metric | Value |
|---|---|
| Servers tracked | 14,824 |
| Categories | 15 |
| Average trust score | 64.5 / 100 |
| Interactions recorded | 101,000+ |
The average trust score of 64.5/100 means the typical MCP server earns a C grade. Not failing, but not something you'd want to blindly trust with your agent's actions.
Distribution by Category
| Category | Servers | Notes |
|---|---|---|
| Code | 5,725 | Largest category by far |
| Other | 3,339 | Uncategorizable servers |
| Search | 1,036 | Brave, Tavily, etc. |
| Data | 974 | Database and data access |
| Productivity | 758 | Calendar, task management |
| Finance | 514 | Payment, trading, accounting |
| Communication | 498 | Email, messaging |
| Media | 393 | Image, video, audio |
| Security | 271 | Auth, scanning, monitoring |
| Education | 235 | Learning, documentation |
Code tools dominate — nearly 39% of all MCP servers.
What Makes a High-Trust Server?
The top-scoring servers share three traits:
- Consistent response times — under 50ms average latency
- Near-perfect success rates — 98%+ of calls returning valid responses
- Active maintenance — recent commits, version updates, responsive to issues
Why This Matters for Agent Builders
If you're building an agent that selects MCP servers dynamically, you need a way to filter out unreliable servers before your agent calls them.
A trust score lets you set a threshold. Below 60? Skip it. Above 80? Proceed with confidence.
Check Any Server in One Line
\bash
curl "https://dominionobservatory.com/api/trust?url=brave-search"
\\
Browse the Full Directory
We launched a browsable directory where you can search, filter by category and grade, and explore server profiles with score breakdowns.
Pre-flight Check
Run a pre-flight check — paste the URL or name, get an instant trust verdict.
Framework Integrations
\bash
pip install fastmcp-dominion
pip install langchain-mcp-trust-gate
pip install crewai-dominion-trust
pip install openai-agents-trust-gate
\\
Each package adds a trust check before every MCP tool call.
Try It
- Directory: dominionobservatory.com/servers/
- Pre-flight Check: dominionobservatory.com/check
- GitHub: github.com/vdineshk/dominion-observatory
No signup needed. No API key required. Free.
I'm Dinesh, building Dominion Observatory in Singapore. If you're building agents that use MCP servers, I'd love to hear what trust signals matter most to you.
Top comments (0)