An MCP server for Clockodo time tracking API. Allows querying, creating, and updating time entries, users, projects, and more for time tracking automation and analysis.
This server has major definition quality gaps. While 16 of 18 tools have descriptions and input schemas are present, many descriptions are generic or lack LLM-optimized guidance. Parameter descriptions are inconsistent, some tools like create_targethour have detailed parameter specs with ranges and constraints, while others like delete lack meaningful parameter descriptions. Tool naming is acceptable but could be more specific (e.g., 'delete' is too generic for a destructive operation). Several tools return raw API responses without documented output schemas, forcing LLMs to parse unstructured data. Error handling is entirely absent, no guidance on what to do when operations fail. The server lacks security-focused descriptions (no mention of which tools require authentication, permission gates, or destructive consequences). Most critically, the two_worktimeschangerequest tools and several others lack complete parameter documentation or have vague type requirements.
Create a new access group.
Create a clock entry (start tracking time).
Create a targethour entry. Specifies how many hours a user is expected to work in a given time period.
Create a worktimes change request to request to modify work times for a user. After creation of the request, the worktime can be changed or deleted. Entries should directly be updated instead of creating a change request.
Delete entity by ID. Dryrun and force only available for customer, subproject, lumpsumservice, project, service.
Stop the clock by ID.
Generic tool name 'delete' with no clarification of what entity type or operation. The description states 'Delete entity by ID' but does not explain which services are destructive, which support dry-run, or what the consequences are. LLMs cannot determine when to use this vs delete_clock or delete_entrygroup.
Tools lacking actionable error guidance. No tool describes what happens on failure (user not found, permission denied, API error). Patterns like 'recovery-guide' and 'error-classification' are absent. LLMs cannot recover from failures, they will retry blindly or give up.
Output schemas not documented. Tools like get_all, get, get_entries, and others return resp.json() with no description of the structure. LLMs cannot plan downstream calls or extract needed IDs without guessing at field names.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 48 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 33 | - | v1 |
Delete entry groups by filter and time range. This can delete a set of work time entries in one call.
Get entity by ID
Get all entities for a given service
Get entries (time entries) for a given time range and optional filters. This gives the actual work time entries, each individual entry. Duration is given in seconds.
Get entry groups (aggregated time entries) for a given time range and grouping. This aggregates the work time entries by the specified grouping criteria and gives the total time for each group. Duration is given in seconds.
Get nonbusinessdays, optionally by year.
Get userreport by ID, year, and type.
Get worktimes change requests with optional filters. Returns a list of change requests, only if a change request has been created before.
Update customer access for a specific access group. If id is provided, updates a specific customer; otherwise, updates general customer access.
Update project access for a specific access group.
Update service access for a specific access group. If id is provided, updates a specific service; otherwise, updates general service access.
Update a targethour entry.
Parameter descriptions are vague or missing context. 'service: Service type to retrieve' does not explain what values are valid (customer, project, user, etc.), forcing the LLM to guess or require extra discovery calls. Enums are not declared.
Destructive operations lack confirmation or dry-run. delete, delete_entrygroup, and delete_clock are destructive but have no mention of rollback, confirmation_key validation, or warning. Pattern 'confirmation-request' not applied.
Composite operations in a single tool. create_clock accepts optional 'away' parameter to set away status, this conflates two concerns (start clock + set away). Should be split into separate tools.
Parameter type 'filter' defined as 'object' with no schema documentation. Tools like get_entries, get_entrygroups, and delete_entrygroup accept 'filter' but do not specify valid filter fields, nesting, or constraints. LLMs cannot construct valid filters.
Pagination incomplete or missing. get_entries and get_entrygroups accept 'page' and 'items_per_page' but responses do not document total_count, has_next_page, or next_page_token. LLMs cannot reliably paginate large result sets.
Secrets and authentication not documented. AUTH_HEADERS are hardcoded from environment variables in clockodo_mcp.py, but tools do not describe authentication requirements or failure modes if credentials are missing. No mention of permission gates or scope declarations.
Idempotency not addressed. Tools like create_accessgroup, create_clock, and create_targethour do not state whether repeated calls with identical inputs return the same result or duplicate records. Agents will retry on ambiguous failures, risking duplicates.