Skip to content

MCP Server

Auditize exposes an MCP server that lets agentic tools such as Claude explore the logs of a repository directly, using natural language.

Connecting to the MCP server

The MCP server is served over streamable HTTP at ${AUDITIZE_URL}/mcp/logs. Two headers are required on every request:

  • Authorization: Bearer ${AUDITIZE_APIKEY}, where ${AUDITIZE_APIKEY} is the secret of an API key with at least read permission on the log repository you want to explore.
  • X-Auditize-Repo: ${AUDITIZE_REPO}, the ID of the repository to explore. The MCP server operates on a single repository at a time.

If the API key is restricted to a subset of log entities, the results returned by the tools are filtered accordingly.

Refer to your agentic tool's own documentation for how to connect it to an MCP server (some tools support remote HTTP servers directly, others require a local bridge).

Available tools

The MCP server exposes the following read-only tools:

  • search_logs and count_logs, to search and count logs using the same filters as the REST API.
  • search_actors, search_resources, search_rich_tags and search_entities, to resolve names into refs usable as filters in search_logs/count_logs.
  • list_action_types, list_action_categories, list_actor_types, list_resource_types, list_attachment_types, list_attachment_mime_types and list_simple_tag_types, to discover the possible values for these filters.
  • list_source_fields, list_details_fields, list_actor_extra_fields, list_resource_extra_fields and their *_field_values counterparts, to discover and filter on custom fields.

Listing and search tools are paginated with cursor-based pagination.