Notion (notion)

The notion connector indexes Notion pages as markdown documents and Notion database (data source) entries as searchable records.

How MFS sees it

notion://workspace/
├── pages/
│   └── <page-id>.md                       document
└── data_sources/
    └── <data-source-id>/
        ├── records.jsonl                  record_collection
        └── schema.json                    table_schema

Pages render to markdown and are searchable with no extra config. Data source records need [[objects]].text_fields to become searchable (their properties are flattened before rendering).

Credentials

You need a Notion Internal Integration token plus explicit page sharing.

1. https://www.notion.so/profile/integrations → New integration → pick the workspace.
2. Under Capabilities, enable Read content (nothing else is needed).
3. Copy the Internal Integration Token (secret_…).

The token alone grants nothing — you must share each page or database with the integration: open the page → ••• → Connections → Connect to → the integration. Sharing propagates to sub-pages, so sharing the workspace home page is the workspace-wide path; sharing one section scopes it there.

Configuration

token = "env:NOTION_TOKEN"

[[objects]]
match = "/data_sources/<data-source-id>"
text_fields = ["Name", "Description"]
locator_fields = ["id"]

The [[objects]] block is only needed for searchable database records — pages work without it.

Sync and freshness

The connector uses each page's last_edited_time as its cursor for incremental re-sync; deletions are caught by full_scan.

Search and browse

mfs add notion://workspace --config ./notion.toml

mfs search "launch checklist" notion://workspace/pages
mfs cat notion://workspace/pages/<page-id>.md --range 1:80
mfs cat notion://workspace/data_sources/<data-source-id>/records.jsonl --locator '{"id":"<page-id>"}'

Pitfalls

• The integration only sees pages and databases explicitly shared with it.
• The plugin uses Notion's data_source API and paths under data_sources/ (older terminology said "databases").
• Database records need text_fields to be searchable.