MCP tool reference
search, get_item, and the conditional get_context tool.
Last verified: 31 July 2026
The server exposes two tools always, and a third when it is enabled for your account. Each runs with your account's permissions and returns only what you can access.
search
Search your indexed content and get back matching items.
| Parameter | Type | Required | Default | Notes |
|---|---|---|---|---|
query | string | Yes | – | What to search for, in natural language. |
limit | integer | No | 10 | Between 1 and 25. |
Results identify each item and give your client enough to decide what is worth reading in full. To read one, pass its id to get_item.
get_item
Fetch one indexed item in full.
| Parameter | Type | Required | Notes |
|---|---|---|---|
id | UUID | Yes | The item id from a prior search result. |
Ids come from search results – they are not stable identifiers you can construct or guess from a provider's own URL.
Capped at 3,000 reads per day.
get_context
Conditional. This tool only appears when the Context Layer is enabled and your account is entitled to it. If your client does not list it, you do not have it.
| Parameter | Type | Required | Default | Range |
|---|---|---|---|---|
task | string | Yes | – | Up to 500 characters. |
limit | integer | No | 12 | 1–25 candidate documents. |
maxTokens | integer | No | 2000 | 200–8000 assembled token budget. |
freshnessDays | integer | No | – | Up to 3650 when supplied. |
get_context returns raw cited evidence, not a written answer. It is a retrieval tool for your client's model to reason over, not a replacement for Ask.
Each response carries:
- The assembled evidence, with citations
- Provenance – where each piece came from
- Freshness – how current the evidence is
- Omissions – what was left out and why
- Warnings – conditions your client should account for
- A token estimate for the assembled context
The final evidence set is often smaller than limit suggests. Candidates are dropped by access filtering, quality filtering, and the token budget, in that order – so asking for 25 documents does not guarantee 25 come back, and a small result is a signal about your corpus, not a failure.
Permissions
Every tool is scoped to your organisation and to your own document permissions – a client sees exactly what you can already see, and nothing beyond it. Within that boundary there is no per-source or read-only limit yet, so a connected client can search and open any document in your own indexed knowledge base. See the MCP overview.
Removed tools
remly.ask and remly.list_recent appeared in earlier documentation. Neither exists on the current server. A client configured against them will fail to call them.