Glean
Arcade MCP server exposing Glean Client API search over custom OAuth2.
Glean toolkit connects Arcade to your organization's Glean enterprise search index via the Glean Client API. It enables AI agents and tools to run permission-filtered, ranked searches over internal company content.
Capabilities
- Enterprise search: Query the authenticated user's Glean index and retrieve ranked, permission-filtered results — users only see content they are authorized to access.
- Pagination: Page through results using
offset+limit;has_next_pagesignals when more results exist. Offset paging is supported up to ~2,000 results (20 pages × 100 items); deeper offsets raise an explicit error. - Result-size safety: When a response would exceed the size budget, the result list is trimmed and
truncatedis set totrue;has_next_pageremainstrueso callers can narrow the query or continue paging. Partial pages are never silently returned — the tool raises an error instead (fail loud over mislead).
OAuth
Authentication uses OAuth 2.0 via a custom Glean OAuth provider. See the Arcade Glean auth provider docs for setup details.
Secrets
GLEAN_API_URL — The base URL for your organization's Glean Client API endpoint. Glean deployments are tenant-specific, so the URL is not shared or discoverable; it takes the form https://<your-tenant>-be.glean.com (or a similar pattern assigned by Glean to your organization). Retrieve it from your Glean admin or from the Glean workspace settings panel. You may also find it documented under your organization's Glean developer or API settings. See Glean's API documentation for context on endpoint structure.
Store this value as an Arcade secret; see Arcade secret configuration docs and manage secrets at https://api.arcade.dev/dashboard/auth/secrets.
Available tools(1)
| Tool name | Description | Secrets | |
|---|---|---|---|
Search the user's Glean enterprise index and return ranked results.
Results are permission-filtered to the authorized user. Use 'offset' with
'limit' to page; 'has_next_page' indicates more results remain. Offset
paging is supported up to roughly 2000 results (20 pages x 100 items);
deeper offsets raise an error rather than return a misleading empty page.
When the fetch budget is exhausted before the requested window is filled,
the tool raises rather than return a partial page (fail loud over mislead).
The result list is trimmed to a size budget when items are large;
'truncated' flags it and 'has_next_page' stays true so the caller can page
or narrow the query. | 1 |