The **guides** surface exposes DailyTube's public documentation (these docs plus the published articles) as searchable, readable data. It powers the `search_docs` and `read_doc` MCP tools and is equally available over plain REST.
```
endpoint: GET /guides/search
endpoint: GET /guides/{document_id}
```
> Guides ships with the developer platform's public beta. Availability follows the API rollout notice on the [MCP page](/docs/mcp).
## Authentication
Guides requires a signed-in identity on every call — either credential works:
- an **API key**: `Authorization: Bearer dtk_YOUR_KEY` — see [Authentication](/docs/authentication)
- an **OAuth access token** issued to an approved MCP/agent client
Unlike account operations, guides deliberately has **no API-enabled plan requirement**: every plan tier, including accounts with no active subscription, may use it. Anonymous calls are rejected.
## Searching
`GET /guides/search?q=<query>&limit=<1..10>` searches full page text as well as titles, descriptions and headings. Results are ranked deterministically and each hit carries a citation: document `id`, `title`, canonical `url`, and a snippet (at most 600 characters).
When nothing matches exactly but a similar title exists, the response returns clearly labeled `suggestions` instead of pretending they were matches. Nonsense queries return an honest empty result — never unrelated filler.
## Reading
`GET /guides/{document_id}` returns the document's content with its citation and the corpus `revision`. Long documents are paginated: when a segment ends mid-document, `nextCursor` is present — pass it as `?cursor=` to continue. Segments are bounded (16,000 characters) so a single response is never unbounded.
An invalid, tampered, or stale cursor (from an older corpus revision) returns `guides.bad_cursor`; an unknown document returns `guides.not_found`.
## Rate limits and logging
Guides calls count against the **same caller budget** as other API operations — API-key calls use that key's budget; OAuth calls share the user's budget across their authorized clients. Every admitted call appears in your [Usage Logs](/usage-logs) with its credential identity. See [Rate limits](/docs/rate-limits).
## MCP tools
Agents connected through MCP get the same capability as two tools:
- **`search_docs`** — the search above, with citations
- **`read_doc`** — the bounded reader above, by document id
Both require an authenticated session and follow the same shared budget. See the [MCP page](/docs/mcp) for connecting an agent.
Guides
The guides surface exposes DailyTube's public documentation (these docs plus the published articles) as searchable, readable data. It powers the search_docs and read_doc MCP tools and is equally available over plain REST.
endpoint: GET /guides/search
endpoint: GET /guides/{document_id}
Guides ships with the developer platform's public beta. Availability follows the API rollout notice on the MCP page (/docs/mcp).
Authentication
Guides requires a signed-in identity on every call — either credential works:
• an API key: Authorization: Bearer dtk_YOUR_KEY — see Authentication (/docs/authentication)
• an OAuth access token issued to an approved MCP/agent client
Unlike account operations, guides deliberately has no API-enabled plan requirement: every plan tier, including accounts with no active subscription, may use it. Anonymous calls are rejected.
Searching
GET /guides/search?q=<query>&limit=<1..10> searches full page text as well as titles, descriptions and headings. Results are ranked deterministically and each hit carries a citation: document id, title, canonical url, and a snippet (at most 600 characters).
When nothing matches exactly but a similar title exists, the response returns clearly labeled suggestions instead of pretending they were matches. Nonsense queries return an honest empty result — never unrelated filler.
Reading
GET /guides/{document_id} returns the document's content with its citation and the corpus revision. Long documents are paginated: when a segment ends mid-document, nextCursor is present — pass it as ?cursor= to continue. Segments are bounded (16,000 characters) so a single response is never unbounded.
An invalid, tampered, or stale cursor (from an older corpus revision) returns guides.bad_cursor; an unknown document returns guides.not_found.
Rate limits and logging
Guides calls count against the same caller budget as other API operations — API-key calls use that key's budget; OAuth calls share the user's budget across their authorized clients. Every admitted call appears in your Usage Logs (/usage-logs) with its credential identity. See Rate limits (/docs/rate-limits).
MCP tools
Agents connected through MCP get the same capability as two tools:
• search_docs — the search above, with citations
• read_doc — the bounded reader above, by document id
Both require an authenticated session and follow the same shared budget. See the MCP page (/docs/mcp) for connecting an agent.
Search and read DailyTube's public documentation through the API — for agents and applications alike.
The guides surface exposes DailyTube's public documentation (these docs plus the published articles) as searchable, readable data. It powers the search_docs and read_doc MCP tools and is equally available over plain REST.
endpoint: GET /guides/search
endpoint: GET /guides/{document_id}
endpoint: GET /guides/search endpoint: GET /guides/{document_id}
Guides ships with the developer platform's public beta. Availability follows the API rollout notice on the MCP page.
Authentication
Guides requires a signed-in identity on every call — either credential works:
an API key: Authorization: Bearer dtk_YOUR_KEY — see Authentication
an OAuth access token issued to an approved MCP/agent client
Unlike account operations, guides deliberately has no API-enabled plan requirement: every plan tier, including accounts with no active subscription, may use it. Anonymous calls are rejected.
Searching
GET /guides/search?q=<query>&limit=<1..10> searches full page text as well as titles, descriptions and headings. Results are ranked deterministically and each hit carries a citation: document id, title, canonical url, and a snippet (at most 600 characters).
When nothing matches exactly but a similar title exists, the response returns clearly labeled suggestions instead of pretending they were matches. Nonsense queries return an honest empty result — never unrelated filler.
Reading
GET /guides/{document_id} returns the document's content with its citation and the corpus revision. Long documents are paginated: when a segment ends mid-document, nextCursor is present — pass it as ?cursor= to continue. Segments are bounded (16,000 characters) so a single response is never unbounded.
An invalid, tampered, or stale cursor (from an older corpus revision) returns guides.bad_cursor; an unknown document returns guides.not_found.
Rate limits and logging
Guides calls count against the same caller budget as other API operations — API-key calls use that key's budget; OAuth calls share the user's budget across their authorized clients. Every admitted call appears in your Usage Logs with its credential identity. See Rate limits.
MCP tools
Agents connected through MCP get the same capability as two tools:
search_docs — the search above, with citations
read_doc — the bounded reader above, by document id
Both require an authenticated session and follow the same shared budget. See the MCP page for connecting an agent.