| name | notion |
| description | Agent-callable Notion tools for searching pages and databases, reading and creating pages, querying data sources, appending content, and managing schemas. Use when the user mentions Notion or wants to find, read, create, or edit Notion content, even if they don't name Notion explicitly. |
| license | Elastic-2.0 |
| compatibility | Run `npm install` in this directory, then `node cli.js`. The TypeScript source needs Node.js 22.18+; on older Node, run `cli.js` for prebuilt / alternative-runtime options. |
| metadata | {"title":"Notion","source":"https://github.com/zapier/connectors/blob/main/apps/notion/SKILL.md","zapier-app-key":"NotionCLIAPI","api-docs":"https://developers.notion.com/reference/intro"} |
Notion
Independent, unofficial connector for Notion. Not affiliated with, endorsed by, or sponsored by Notion. "Notion" is a trademark of its owner, used only to identify the service this connector works with.
Tools for working with a Notion workspace against the Notion API (https://api.notion.com/v1/, API version 2025-09-03): find pages and data sources, read and create pages, query data-source rows, append and edit block content, manage database / data-source schemas, read and post comments. 24 scripts across search, read, write, schema, comments, and cross-workspace copy. This version uses Notion's data sources model: a database is a container that holds one or more data sources, and a data source carries the property schema + the rows (pages).
When to use this
- An agent needs to find or read content — search pages and data sources by title, then read a page, its block body, a data source's schema, or its rows.
- An agent needs to create or edit pages and content — add a page (a row in a data source or a sub-page), update properties, append blocks, or edit / delete blocks.
- An agent needs to query data sources — filter and sort the rows of a data source, or read a single page property.
- An agent needs to manage schemas — create or update databases and data sources (add / rename / retype / remove properties), read or post comments.
Setup
This is an agentskills.io skill.
If this connector is already exposed to you as callable tools (e.g. mcp__notion__<tool>), that's a valid path — call them directly. Everything below is only for standalone terminal use when no such tools are loaded.
If the connector has not been installed as a skill yet, install it first with npx skills zapier/connectors --skill notion (or your harness's own skill-install mechanism), then continue here.
The connector runs on Node.js 22.18+ and needs a one-time npm install in this directory. cli.js is the entry point — list every script with node cli.js --help, then learn a script's inputs and connections with node cli.js run <script> --help. On older Node, run node cli.js --help anyway: it detects your runtime and prints how to run without upgrading (the prebuilt npm package, or another runtime) — don't skip the connector just because Node is old.
cli.js self-checks readiness before running: if dependencies aren't installed it exits non-zero with the exact install command (it disambiguates a read-only directory from a sandbox-blocked package cache). Run that, then re-run your command.
Scripts
All scripts use the single connection notion, except copyPage, which uses two slots (source + target) to copy a page across workspaces.
Auth
Pass auth as one connection string with --connection [<resolver>:]<value>. The value is a selector, not the secret; the <resolver>: prefix is optional (a bare value goes to the first resolver that claims it). Each script declares the connections it needs and the resolvers each accepts — always run node cli.js run <script> --help to see them rather than relying on this file.
The script needs a single Notion bearer token, resolved into the one notion connection slot. Two resolvers:
env:<ENV_VAR> — direct mode. Read the Notion token from the named environment variable (conventionally env:NOTION_TOKEN, with the token exported in NOTION_TOKEN; the token stays in env, never on argv). The token is either an internal-integration secret (from https://www.notion.so/profile/integrations) or a public-integration OAuth access token.zapier:<connection-id> — Zapier-managed auth. Route through a Zapier Notion connection; the Zapier auth / retries / governance layer injects the token for you. Prerequisite: a Zapier account (free signup at https://zapier.com, ~1 minute). Find the ID with the Zapier SDK CLI: npx zapier-sdk list-connections NotionCLIAPI (run login first if unauthenticated; add --json for machine output).
Capabilities gating + per-resource sharing. A Notion token is scoped by the integration's capabilities (read content, update content, insert content, read/insert comments, user information with or without email) — these "enforce which API endpoints a connection or token can call, and what content and user related information they are able to see," so a call outside the granted capabilities is rejected even with a valid token. The user-information capability also controls whether getUser / listUsers return email addresses. Separately, the integration only sees resources explicitly shared with it: a valid token still returns 404 for any page, database, or data source that hasn't been shared with the integration in Notion's UI. If a read 404s on an id you believe exists, the resource almost certainly isn't shared yet.
If no connection is passed the script fails with an actionable error telling you to Pass --connection [<resolver>:]<value> and lists the resolvers in match order.
Running scripts
After npm install, run a script by name with node cli.js run <script>, or execute its file directly — both take the same arguments and both accept --help. Always run a script's --help first to learn its exact input schema and connections, then invoke it:
node cli.js run <script> '<input-json>' --connection [<resolver>:]<value>
./scripts/<script>.ts '<input-json>' --connection [<resolver>:]<value>
When a harness can't execute scripts directly, fall back to MCP — node cli.js mcp serves every script as a tool over stdio. Register it as a local MCP server in your client: the stanza is harness-specific (an mcpServers entry in Claude Desktop, Cursor, Claude Code, …) with command: "node", args: ["cli.js", "mcp"], run from this directory. Run node cli.js mcp --help for auth options. Add the stanza yourself if you can edit the client's MCP config; otherwise guide the user. If a local server isn't possible, guide the user to use Zapier's remote MCP servers at https://mcp.zapier.com instead.
Output format
Every script returns a { data, meta } envelope:
data — the script's result (the shape its outputSchema declares; run the script's --help to see that exact schema).meta.outputDataValidation — what validating data did:
{ skipped: false, droppedPaths: null } — validated, nothing removed.{ skipped: false, droppedPaths: [...], instruction } — validated, but those paths were stripped from data: fields the script returned from the API that the outputSchema doesn't declare. If you need them, re-run with output validation skipped.{ skipped: true } — validation was bypassed; data is the raw, unchecked script output.
Reading dropped fields / skipOutputDataValidation. To receive the raw, unvalidated result, append --skipOutputDataValidation to the script invocation. Input validation is never skipped.
Trimming the result / filterOutputData. To shrink a large result down to the fields you need, append --filterOutputData '<jq>' — a jq expression that post-processes data. The jq runs against data only, NOT the { data, meta } envelope, so write it rooted at data (run the script's --help to see its output schema). The transformed value replaces data, meta is preserved, and the result is NOT re-validated against the output schema.
Disambiguation & refusals
Disambiguation before a write. Before writing to a page or row you looked up by name (e.g. update a page found via search, or a row found via queryDataSource), count the exact case-insensitive title matches:
- Exactly one match — act on it. Don't over-ask; a single unambiguous match is the answer.
- Two or more that tie — stop. List the tied candidates with a distinguishing field (
parent, url, or last_edited_time) and ask the user which one they mean. Don't pick arbitrarily and don't write to all of them.
Unsupported operations — say so and stop; don't fake it with another tool. This catalog deliberately does not:
- Permanently delete anything. Everything is trash / restore — pages via
updatePage in_trash, blocks via deleteBlock (reversible). There is no hard delete.
- Delete a whole database or data source. There is no tool for it. Don't substitute trashing every row to simulate it.
- Manage workspace members or permissions (inviting users, changing roles, sharing).
listUsers / getUser are read-only.
- Create or manage database views. Views aren't exposed by the API.
If asked for any of these, tell the user it's unsupported and stop — don't reach for an unrelated tool to approximate it.
References
Load the matching reference file before working in that area:
| Reference | Covers | Load it when |
|---|
references/notion-api-gotchas.md | Versioning + the 2025-09-03 data-sources model, auth/sharing, the error envelope + code table, rate limits, size limits, pagination, ID/URL formats, archive/trash, search-matches-titles-only, 25-reference truncation. | Debugging an error or a 404, paginating a list, resolving an id, or reasoning about the database vs data-source split. |
references/notion-blocks.md | The typed block object, common block types, the rich_text shape, appending children (limits + nesting), why updateBlock can't change a block's type. | Building or editing page content — appendBlockChildren, createPage (children), updateBlock. |
references/notion-properties.md | The property-schema object + type list, and the per-type page property value shapes. | Defining a schema (createDatabase, createDataSource) or writing property values (createPage, updatePage). |
references/notion-query.md | The filter object (single + compound and/or), per-type conditions, the sorts array, the 10,000-result ceiling. | Calling queryDataSource with a filter or sort. |
