# Give your agent the answer, not the query language

DataFocal is an MCP server. Your agent asks what changed and why, and gets findings with the evidence attached. Queries are never metered, so an agent can run thousands in one investigation.

## Connect in one line

Claude Code:

```
claude mcp add --transport http datafocal http://localhost:8080/mcp
```

Codex:

```
codex mcp add datafocal --url http://localhost:8080/mcp
```

Any MCP client:

```
{ "datafocal": { "url": "http://localhost:8080/mcp" } }
```

## What your agent can ask

- "Where does our agent spend go?": Spend by model, tool, repo and run, with savings ranked in dollars a day.
- "Why did that run fail?": The turn where it went wrong, the tool calls around it, and similar runs.
- "What changed since the deploy?": Findings ranked by how unusual they are, each with evidence.
- "Drill into checkout errors": Narrow to a slice and see what stands out at the next level.
- "Show me the raw events": The exact records behind any number, trimmed to fit the context window.
- "Are we within our SLOs?": Status and error budget for every objective you define.

## DataFocal at a glance

- Product: DataFocal: AI observability, then software and edge observability
- What it does: Shows what AI agents and LLM apps are doing: spend, cost per run, cache hit rate, failures and what changed. Explains it in plain language and drills down to the runs behind it.
- Data: Claude Code and Codex transcripts, LLM gateway logs, OpenTelemetry GenAI traces, application logs and traces, Kubernetes, robot and edge device events
- Pricing: Open source free; Team $30 per user per month plus collectors from $12 per month (free on developer machines); Enterprise custom. Storage managed at cost + 20% or $0 in your own bucket. No charge for data volume or queries
- Open source: Apache 2.0 license, github.com/datafocal/datafocal
- Agent access: MCP server, http://localhost:8080/mcp on a local install
- Contact: sales@datafocal.ai