Typefully API integration MCP server for Cursor - provides tools to create, manage, and retrieve drafts from Typefully
This server has moderate structural definition quality but significant gaps in descriptions, parameter documentation, and error handling. Tool names are clear and action-oriented (list_, create_, get_), and schemas are present with type information. However, parameter descriptions are minimal or missing context, several tools lack actionable error guidance, and output schemas are generic JSON objects without field-level documentation. The server defines 4 tools with schemas visible in code; however, 'list_tools' is a meta-tool (self-referential discovery) which is an antipattern, MCP servers should expose tools via the native protocol, not via a custom tool. The typefully_* tools have basic structure but lack production-grade completeness.
Returns a JSON list of all available tools with their descriptions, parameters, and examples
Creates a new draft in Typefully
Retrieves recently published drafts from Typefully
Retrieves recently scheduled drafts from Typefully
list_tools meta-tool is an antipattern. MCP servers natively expose tools via ListTools; a custom 'list_tools' tool duplicates protocol-level discovery and indicates misunderstanding of the MCP design model.
Parameter descriptions lack actionable context. E.g., 'Whether to threadify the content (default: false)' does not explain what threadifying means, when to use it, or how it affects output. LLMs cannot infer; they need explicit guidance.
List-returning tools lack pagination parameters (page, offset, limit) and result limits. typefully_get_scheduled_drafts and typefully_get_published_drafts could return hundreds of items, blowing context windows. No documentation of result count or truncation behavior.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 51 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 38 | - | v1 |
Output schemas are overly generic. All tools return a loose 'object' with minimal property constraints. No field-level type information, nullability, or structure. Example: create_draft response schema lists 'data: { type: "object" }' with no nested field definitions. LLMs cannot extract or chain results reliably.
Error handling is minimal and non-actionable. Code catches errors generically and returns 'Failed to create Typefully draft' with error details, but does not classify errors as retryable, user-fixable, or fatal, nor does it suggest recovery steps (e.g., 'Check API key' or 'Retry with different schedule date'). No dry-run or confirmation for destructive operations like draft creation.
No tool accepts human-friendly identifiers or performs lookups. For example, if an agent wants to attach a draft to a campaign or user, the tool likely requires an opaque ID. No patterns for accepting names/emails and resolving them internally, a violation of the chat-data-model principle.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) present. The MCP spec and production baselines expect tools to declare their side effects. create_draft is destructive but unmarked; idempotency is undefined.
Parameter 'schedule_date' is described as ISO format but has no validation or error guidance. If an agent passes an invalid date, the tool error will not tell it why or how to fix it. Should document format strictly (e.g., 'ISO 8601 instant, e.g., 2025-02-10T15:30:00Z').