Author
Mano Toth
Senior Technical Writer
If you don’t need a blog post to convince you that connecting with Axiom via Model Context Protocol (MCP) would be neat, the set-up instructions are right here.
Honestly, we recommend you try it first! But if you haven’t got an Axiom account or an MCP client near to hand, you can see an example of the Claude desktop app interfacing with Axiom using the Axiom MCP server in the video below:
Whether you’ve tried it out on your own data, or just watched the video, hopefully it’s now clear — with Axiom’s MCP server, your agents gain direct access to your organization’s event data, within the boundaries that you define through access control. They can query, analyze, and reason through complex debugging scenarios without you having to manually craft queries or dig through dashboards.
More context, please
Debugging production issues is fundamentally about testing hypotheses. Is it a deployment? A traffic spike? A dependency failure? The traditional approach to leveraging LLMs for debugging requires guessing at the right context, dumping it into a prompt and hoping for the best. That works fine for simple questions, but debugging complex distributed systems requires iterating through dozens of potential causes.
MCP flips this around. Instead of relying on the right context making it into the prompt, agents decide what they need to know and go get it. They can test multiple hypotheses in parallel, correlate data across different time windows, and drill down into specific traces when patterns emerge.
Axiom’s APL query language is particularly well-suited to this approach. You can use APL to slice and dice event data in any way you need, but building a complex query can often take time, and during an incident time is a scarce resource. The good news is that agents excel at leveraging APL to its full potential. They know every attribute in your dataset, every aggregation we support, and can easily rule out a group of hypotheses with a single query.
The result is that you’re letting the AI handle more of the problem. Instead of you figuring out what to ask and how to ask it, you just describe the issue and let the agent determine the investigation path.
You aren’t sending us enough events
Most organizations only send logs from some of their systems, and only scope access to some of their teams. This makes it difficult to leverage the hidden insights in your event data that rely on cross-functional correlations.
Consider this scenario: checkout completion rates drop in your mobile app, but only for users in specific geographic regions. With traditional monitoring, the payments team sees declined transactions, the mobile team sees app crashes, and the infrastructure team sees elevated error rates. Each team investigates their piece of the puzzle separately.
With an agent that has access to the event data that you define, the investigation connects these dots automatically. It correlates the payment declines with specific app versions, maps the crashes to CDN performance in those regions, and identifies that a recent change to your edge caching strategy is causing checkout flows to timeout. What would have taken three teams days to coordinate becomes a single investigation that completes in minutes.
Event data is extraordinarily rich when you can correlate it properly. The challenge has always been that correlation is hard — until agents make it trivial. The organizations that embrace sending all their event data to a single platform, accessible by agents, are the ones that will debug faster, understand their systems better, and ultimately ship more reliable software.
Context is what makes organizations unique. It’s the accumulated knowledge of how your specific systems behave, what business as usual looks like for your workloads, and where the problems arose in previous incidents. Organizations that figure out how to make this context accessible to their agents will have a significant advantage over those that don’t.
Connecting safely
Of course, giving agents access to all your data requires careful controls. Axiom’s role-based access control (RBAC) ensures agents only see what they’re supposed to see. If your payments team shouldn’t access infrastructure logs, neither should an agent running on behalf of that team.
The MCP server currently supports read-only operations, so agents can investigate but can’t modify your data or configuration. Every query is logged in Axiom’s audit trail, giving you complete visibility into what agents are doing with your data.
The final key ingredient is Axiom’s hyper-efficient query engine, which ensures that thousands of complex, agent-driven queries remain cost-effective. For complete control and peace of mind, we’ve also implemented spend controls that allow you to set monthly limits on usage, avoiding any billing surprises.
What’s next
Axiom is becoming the hub of organizational context for our customers. All event data, all the time, being queried continuously by agents working to understand and improve your systems.
The goal isn’t to replace human intuition in debugging — it’s to amplify it. Agents handle the mechanical work of querying and correlating data, while you focus on the creative work of understanding what it all means and deciding what to do about it.
Ready to give your agents the context they need? Check out the MCP server documentation and start using your event data to its full potential.
