- LATENCY
- ERRORS
- PROJECTS
- HUMAN
Short answer
How can I use AI to debug ML execution spans?
Your AI pulls trace data from Arize AI to find latency or errors in your model runs. It scans your project history to highlight the exact spans causing issues. You get a clear view of the bottlenecks.
Where your data lands.
Debugging outcomes.
The AI organizes trace information into actionable categories.
LATENCY
Slow execution spans
The AI identifies specific spans with high duration. You get a list of the exact steps dragging down your performance.
ERRORS
Failed model runs
The AI finds spans that returned errors. It isolates the specific trace IDs that need your attention.
PROJECTS
Active ML projects
The AI lists your available projects. This lets you narrow your search to a specific model or deployment.
HUMAN
Manual code fixes
The AI points out the problematic spans, but you still need to jump into your codebase to apply the actual logic changes.
The workflow
What your AI does when the request arrives.
The AI moves from broad project oversight to granular trace inspection.
Identify the project
The AI looks at your available ML workloads to find the right context for your query.
list_projectsLocate specific traces
Once the project is set, the AI pulls the individual execution spans you asked about.
list_spansAnalyze span data
The AI reviews the timing and status of the retrieved spans to find outliers.
list_spansSummarize bottlenecks
The AI presents a clear explanation of which spans are causing the most trouble.
list_spans
Try it
Copy these to start.
Use these prompts to begin your debugging session.
Starting points
Swap out the project names and span IDs with your actual Arize AI identifiers.
Arize AI Connector
You're all set. Choose your MCP client and follow the setup instructions.
https://edge.vinkius.com/vk_preview_wKJSgGsb2EgJmks5PzM3V0jknb7nFZgbwB0CaRFP/mcpClaude Desktop
Follow the steps below to connect in seconds.
- 1In Claude Desktop, open Settings → Connectors.
- 2Click “Add custom connector” and paste the connector link above as the remote MCP server URL.
- 3Click Add and start a new chat — Arize AI capabilities are ready to use.
{
"mcpServers": {
"arize-ai-mcp": {
"url": "https://edge.vinkius.com/vk_preview_wKJSgGsb2EgJmks5PzM3V0jknb7nFZgbwB0CaRFP/mcp"
}
}
}List all my ML projects in Arize AI.
Find the execution spans for my 'recommendation-engine' project.
Which spans in my current project are taking the longest to execute?
Show me the most recent error spans from the 'image-classifier' project.
Claude
ChatGPT
Cursor
VS Code
Windsurf
Claude Code
JetBrains
Cline
Start here
Connect Arize AI once, then ask.
Just link your account once. Your credentials stay encrypted, and you can immediately start asking your AI about your traces.
Connect Arize AI to your AIFAQ
How this task behaves.
- 01
Can the AI fix my model code directly?
No. The AI can only read trace data and identify where the problems are. You have to make the code changes yourself.
- 02
Can the AI delete old traces in Arize AI?
No. The tools only allow the AI to list projects and spans. It cannot modify or delete any data in your Arize account.
- 03
How does the AI know which project to look at?
It uses the list_projects tool to see what you have available, then you tell it which one to inspect.
- 04
Does the AI see all my data?
The AI only sees the project and span information it needs to answer your specific questions.
- 05
Can I use this with Cursor or Claude?
Yes, as long as you are using an MCP-compatible client, you can use this Connector.
More questions about Arize AI? The Connector page answers them. See everything the Arize AI Connector can do
Connect Arize AI to Claude, Cursor, ChatGPT & more
