Context
Benchmarking against mole's MCP server implementation and standard MCP best practices.
Problem
autotask-mcp currently implements tools and resources, but registers 0 MCP Prompts (s.AddPrompt).
Autotask PSA is heavily process- and workflow-driven (e.g. ticket triage, incident post-mortems, billing summary preparation, daily timesheet review). Without pre-defined MCP prompts, AI assistants must construct ad-hoc instructions from scratch each time, leading to inconsistent outputs and missed metadata (such as failing to check queue IDs, priority levels, or billable role IDs).
Proposed Solution
Register standard MCP Prompts in a new prompts/ package:
autotask_triage_ticket:
- Arguments:
ticketId (int64) or description (string)
- Fetches the ticket, related company, open tickets for the same company, and picklists for queues/priorities. Prompts the model to categorize the issue, select the optimal queue, set priority, and draft an initial response or internal note.
autotask_summarize_ticket:
- Arguments:
ticketId (int64)
- Aggregates ticket details, recent notes, and time entries to generate a concise status summary for handoffs or customer updates.
autotask_draft_time_entry:
- Arguments:
ticketId (int64), hoursWorked (float64), summary (string)
- Guides the assistant to resolve the correct role ID, verify billable status, format the work summary note, and call
autotask_create_time_entry.
autotask_weekly_timesheet_review:
- Arguments:
resourceId (int64), startDate (string), endDate (string)
- Inspects logged time entries across tickets and projects to check for missing hours, unsubmitted entries, or mismatched billing codes.
Acceptance Criteria
s.AddPrompt is used to register prompts with descriptions and argument schemas.- Prompts return structured messages that guide LLM interactions through standard PSA workflows.
- Registered prompts are verified via unit/integration tests with in-memory MCP clients.
Context
Benchmarking against
mole's MCP server implementation and standard MCP best practices.Problem
autotask-mcpcurrently implements tools and resources, but registers 0 MCP Prompts (s.AddPrompt).Autotask PSA is heavily process- and workflow-driven (e.g. ticket triage, incident post-mortems, billing summary preparation, daily timesheet review). Without pre-defined MCP prompts, AI assistants must construct ad-hoc instructions from scratch each time, leading to inconsistent outputs and missed metadata (such as failing to check queue IDs, priority levels, or billable role IDs).
Proposed Solution
Register standard MCP Prompts in a new
prompts/package:autotask_triage_ticket:ticketId(int64) ordescription(string)autotask_summarize_ticket:ticketId(int64)autotask_draft_time_entry:ticketId(int64),hoursWorked(float64),summary(string)autotask_create_time_entry.autotask_weekly_timesheet_review:resourceId(int64),startDate(string),endDate(string)Acceptance Criteria
s.AddPromptis used to register prompts with descriptions and argument schemas.