A specialized Model Context Protocol (MCP) server that integrates Notion into your AI workflows. This server enables seamless access to Notion through MCP, allowing AI agents to interact with pages, databases, and comments.
This server demonstrates moderate quality with consistent tool naming and explicit JSON schemas for all 8 tools. However, several critical gaps limit production readiness: (1) output schemas are entirely undocumented, LLMs cannot infer what fields to expect from tool responses; (2) parameter descriptions exist but lack critical detail (no min/max bounds, no format constraints, no enum alternatives where appropriate); (3) error handling is absent, no guidance for recovery, retries, or failure modes; (4) parameter validation is not visible in the provided code; (5) the sampling feature (tools 6-7) uses deprecated MCP patterns (_meta.sampling, explicit temperature/maxTokens) that should be removed or migrated to current Spec; (6) no tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk levels. Strengths: all tools follow verb_noun naming (list_, search_, get_, create_, update_, delete_); descriptions are concrete and explain WHAT the tool does; input schemas are well-formed JSON Schema with additionalProperties: false; risk levels are labeled. The server is buildable and uses current SDK version (0.6.0), but lacks the polish expected for production integration with hosted MCP clients.
Creates a rich, comprehensive Notion page that expands upon basic user inputs. Takes simple instructions and content, then generates a detailed, well-structured page with appropriate sections, formatting, and supplementary content.
Permanently deletes a specified Notion page and all its contents. This action cannot be undone, so use with caution.
Retrieves comprehensive details of a specific Notion page, including its content, properties, and metadata. Returns the complete page structure and all nested content blocks.
Lists all accessible Notion databases in your workspace, sorted by last edited time. Returns key metadata including database title, schema, and last edited timestamp.
Lists all accessible Notion pages in your workspace, sorted by last edited time. Returns key metadata including title, URL, and last edited timestamp.
No output schemas documented. LLMs cannot infer response field names, types, or structure. Responses appear to be unstructured (potentially raw API responses). This blocks downstream tool composition and requires LLMs to parse responses as free text.
Parameter descriptions lack critical detail. 'maxResults' has no bounds (what is the minimum? maximum? is there a cost to large values?). 'userInstructions' is a freeform string with no length limit or format guidance. 'pageId' and 'databaseId' have no format or length specification. This invites invalid LLM inputs and silent failures.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 53 | <=2025-11-25 | v2 |
| 2026-03-09 | D | 58 | - | v1 |
Performs a full-text search across all accessible Notion pages using the provided query. Searches through titles, content, and metadata to find relevant matches.
Searches specifically for Notion pages with titles matching the provided query. Useful for finding exact or similar title matches when you know the page name.
Updates an existing Notion page with rich, comprehensive content based on user instructions. Takes simple inputs and transforms them into well-structured, detailed page content while preserving existing information. Can enhance, reorganize, or expand the current content while maintaining page integrity.
No error handling guidance. Tools lack descriptions of: (1) what errors can occur (e.g., 'database not found', 'permission denied', 'rate limited'); (2) which errors are retryable; (3) what the LLM should do next (e.g., 'try search_notion_pages_by_title if ID lookup fails'). Agents cannot recover gracefully from failures.
Deprecated sampling pattern used in _meta for tools 6-7 (systemprompt_create_notion_page, systemprompt_update_notion_page). The _meta.sampling field with explicit prompt/maxTokens/temperature is deprecated in MCP spec 2026-07-28. This pattern will be removed ~2027-07-28. Migrate: remove _meta.sampling and integrate sampling logic directly with the LLM provider API, or use official extensions.
No tool annotations despite clear risk levels. Tools are labeled READ_ONLY, WRITE, REVERSIBLE, DESTRUCTIVE in comments, but MCP schema lacks readOnlyHint, destructiveHint, idempotentHint annotations. Clients cannot distinguish safe from dangerous tools without parsing descriptions.
Destructive tool (systemprompt_delete_notion_page) lacks confirmation/dry-run support. Agents can permanently delete pages with one invocation. No mechanism to require user approval or preview impact before execution.
List tools (systemprompt_list_notion_pages, systemprompt_list_notion_databases) have no pagination support beyond maxResults. No 'offset', 'cursor', or 'next_token' parameter documented. Large workspaces could exhaust context window or miss results.
Parameter names inconsistent with Notion API. 'databaseId' and 'pageId' are JavaScript camelCase, but Notion's official SDK and docs use underscores (database_id, page_id). If LLMs have prior knowledge of Notion, this mismatch causes confusion. Standardize on snake_case to match official APIs and improve composability.