The Tugboat MCP server has 12 tools with basic naming and structure, but significant gaps in schema completeness, parameter documentation, and output specification. All tool descriptions are present but quite brief (10-50 chars). Most critically, input schemas are not visible in the provided source code, only summary descriptions are given. No output schemas are documented. Error handling and recovery guidance are absent. The server follows a verb_noun naming convention consistently (createPreview, buildPreview, etc.), which is good, but lacks the rigor expected for production use. This is typical C-grade MCP server work.
Start a build for a preview
Create a new preview in a repository
Create a new project
Delete a preview
Delete a project
Get logs for a preview
Retrieve details about a specific project
Input schemas not visible in source code, only high-level parameter descriptions provided.
Output schemas are not documented. LLMs cannot plan downstream tool calls or extract required fields without knowing what fields are returned by each tool.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 42 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 35 | - | v1 |
List all projects with optional pagination
Start a refresh for a preview
Search for previews by name or other criteria
Search for projects by name or description
Update an existing project
Tool descriptions are very brief (10 - 50 characters) and lack context on WHEN to use each tool, WHAT it returns, and prerequisites. For example, 'Start a build for a preview' doesn't explain what a 'build' entails or what state the preview must be in.
Destructive tools (deletePreview, deleteProject) lack confirmation or dry-run capability. No error guidance tells an LLM what to do if deletion fails or what the consequences are.
No error handling or recovery guidance documented. When a tool fails, there are no hints for the LLM on whether to retry, ask the user, or call a different tool.
Tool composition and chaining not documented. For example, if createPreview returns a previewId, is it guaranteed to match the input expected by buildPreview? Output fields that enable tool chaining are not specified.
listProjects accepts limit and offset but no documentation of what fields are returned, how many results are typical, or whether pagination is guaranteed to be consistent. Result capping not documented.
Parameter constraints not stated in descriptions. For example, createProject's 'name' parameter has no length limits, character restrictions, or uniqueness rules documented. LLMs will guess.
API key is passed to TugboatApiClient but exposure mechanism unclear. If credentials are passed as parameters to tools, they will leak into logs and prompt history.