Axiom MCP Server

The Axiom MCP server connects AI assistants to Axiom, an observability and log analysis platform. Anyone collecting logs, traces, or metrics of their applications in Axiom can ask the assistant questions like "Why have error rates been rising since the last deployment?" or "Show me the slowest requests of the last hour". The assistant translates the question into a query in the Axiom Processing Language (APL), runs it, and evaluates the result in the conversation. The server is operated by Axiom itself and, according to the vendor, is the recommended way to connect Axiom to assistants.

What is the Axiom MCP server?

The server implements the Model Context Protocol (MCP), an open standard through which AI applications talk to tools and data sources. The recommended variant is a hosted remote server at https://mcp.axiom.co/mcp; for older clients that only support HTTP with server-sent events, the documentation additionally lists https://mcp.axiom.co/sse. A local installation via go install is described as well, but it needs an API token instead of the browser sign-in, and the prebuilt prompts are not available there. The older, separate project mcp-server-axiom is deprecated according to the vendor and is not covered here.

What the assistant can do with it

According to the documentation the server provides tools for querying data, for exploring datasets and metrics, and for managing dashboards, monitors, and notifiers. Prebuilt prompts for common investigations come on top. The complete tool list lives in a separate reference at Axiom and was not examined item by item for this description. In practice that means the assistant can first find out which datasets and fields exist, then query specifically, and suggest a dashboard or monitor when it spots a pattern. That saves switching between chat, console, and query editor.

Setup and sign-in

You need an Axiom account and a dataset that receives data. For Claude Code a single command with the HTTP transport is enough according to the documentation; in Claude itself the server can be added as a custom connector, on team and enterprise plans possibly first by an administrator. Cursor takes the address through its MCP configuration or an install link. Sign-in runs via OAuth in the browser. Alternatively the documentation names a personal access token as an Authorization header together with the organization ID. You can end access later in the Axiom profile settings by deleting the session.

Security and limits

Because the tools also include management functions for dashboards, monitors, and notifiers, you should not assume read-only access. The documentation says nothing explicit about this; the actual permissions follow from your account or token. Grant tokens only the permissions that are needed. Log data often contains user data, addresses, or error messages with internal details, and query results enter the language model's context and thereby reach the model provider. According to the documentation remote queries are routed through infrastructure in the United States, which can matter for data protection requirements. Axiom also lets you set cost limits for queries; that is worth doing because an assistant could otherwise issue expensive queries over large time ranges.

Costs

The documentation does not bill the MCP server separately. Axiom offers a permanently free Personal plan for individual developers with a monthly free allowance and limited retention, above it the paid Axiom Cloud plan with a platform fee and usage-based billing, and individual enterprise terms via sales. Queries by the assistant count as normal usage. Current terms are on the official pricing page.

Who is the Axiom MCP server for?

It is worthwhile for development and operations teams that already use Axiom and want to speed up troubleshooting, incident analysis, or monitor maintenance through conversation. It is especially useful when APL is not familiar to everyone on the team, since the assistant writes the queries. Anyone not using Axiom gains nothing: the server is an access path to the platform and not a standalone log analysis.

An Axiom account with at least one dataset and an MCP client that supports remote servers.

The Axiom MCP server runs hosted by Axiom. This guide walks through setup, sign-in, and the first queries.

Step 1: Prerequisites

You need an Axiom account, a dataset receiving data, and an MCP client that supports remote servers.

Step 2: Set up in Claude Code

Run claude mcp add --transport http axiom https://mcp.axiom.co/mcp and then call /mcp in Claude Code to complete the browser sign-in. For older clients that only handle HTTP with server-sent events, the documentation lists the address https://mcp.axiom.co/sse.

claude mcp add --transport http axiom https://mcp.axiom.co/mcp

Step 3: Set up in Cursor or Claude

In Cursor enter https://mcp.axiom.co/mcp as the url in the MCP configuration or use the install link from the documentation. In Claude add a custom connector with the same address; on team and enterprise plans an administrator may need to release the connector first.

Step 4: Sign in with a token

If OAuth is not possible, you can pass a personal access token as an Authorization: Bearer header and the organization ID as the header x-axiom-org-id. Give the token only the permissions it needs.

Limit costs and permissions

Set cost limits for queries in Axiom so the assistant does not run expensive queries over large time ranges. Keep in mind that the server can also manage dashboards, monitors, and notifiers.

Step 5: First queries

Let the assistant list the existing datasets first. Then ask, for example, for the most frequent error messages of the last hour and check the generated APL query before you trust results.

OAuth via browser sign-in (recommended) or a personal access token as a bearer header with the organization ID; an API token for the local variant.

Querying and exploring datasets and managing dashboards, monitors, and notifiers; the scope follows the permissions of the account or token. The documentation does not state a read-only mode.

The server runs hosted by Axiom; according to the documentation remote queries run through US infrastructure. Query results enter the AI client context and therefore reach the model provider.

Write-capable management functions for dashboards and monitors, personal data in logs reaching the model provider, overly broad token permissions, and expensive queries without a cost limit.