A MCP server for website tracking data from Umami Analytics
The server provides 4 well-named analytics tools with reasonable descriptions and documented input schemas. However, there are significant gaps in error handling, output schema documentation, and parameter type specificity. Tools follow the verb_noun convention (get_*) appropriately. Descriptions are present but inconsistent in depth, some exceed the optimal 10-200 char range. Parameter descriptions are generally adequate but lack explicit format/constraint details. No output schemas are documented, forcing the LLM to infer result structure. Error handling is minimal, tools note timestamp validation issues but provide no recovery guidance. Overall, the server is functional and above-average for community standards but lacks production-grade polish.
Get a list of the unique session IDs who visited a specific website within a time range and perform a specific event. WARNING: due to api limitations, only the first 1000 total session IDs will be returned by the api. Within those less will be unique. Do not use this tool to calculate the number of unique visitors - only use it to get session IDs. Note: If no results are returned, do not immediately assume there is no data - verify the unix timestamps are correct and ask the user for specific date ranges if not provided.
Get the user journey for a specific session ID within a time range. Note: If no results are returned, do not immediately assume there is no data - verify the unix timestamps are correct and ask the user for specific date ranges if not provided.
Get the 5 overivew metrics for a specific website within a time range. The returned metrics are as follows: - pageviews: The number of total pageviews for the entire website - visitors: The number of unique visitors the website has had - visits: The number of unique visits those visitors have had to the website - bounces: The number of visitors that left the website without interacting with it - totaltime: The total time spent on the website by all visitors Note: If no results are returned, do not immediately assume there is no data - verify the unix timestamps are correct and ask the user for specific date ranges if not provided.
No output schemas documented for any tool. LLMs cannot infer result structure, field names, or data types. Downstream tool composition is error-prone without knowing what fields are available.
Error handling is minimal. Tools mention timestamp validation issues in descriptions but provide no recovery guidance. No error categorization (retryable vs user-fixable vs fatal). LLMs cannot determine whether to retry, ask user, or abort.
Date format descriptions are verbose (exceeding 200 char optimal range) and include examples that LLMs may reuse literally. Should use a concise description + format constraint in schema instead.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 54 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 43 | - | v1 |
Retrieve a list of the websites present in the tracking database. This tool does not require any input. The output of this tool includes the following fields for each website: - id: The unique identifier of the website - name: The name of the website - domain: The URL of the website - shareId: The unique identifier that can connect websites together - resetAt: The date and time when the website was last reset - userId: The unique identifier of the user that owns the website - teamId: The unique identifier of the team that owns the website - createdBy: The unique identifier of the user that created the website - createdAt: The date and time when the website was created - updatedAt: The date and time when the website was last updated - deletedAt: The date and time when the website was deleted - createUser: The unique identifier of the user that created the website, and their username
get_session_ids description warns about API limit (first 1000) but provides no pagination mechanism in the tool itself. The helper function implements pagination internally but the tool schema does not expose page/limit parameters to the LLM, limiting discoverability and control.
get_website_stats description contains a typo ('overivew' instead of 'overview') and is lengthy (over 400 chars). Should be condensed and spell-checked. Descriptions over 200 chars waste tokens.