> For the complete documentation index, see [llms.txt](https://docs.katara.ai/about/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.katara.ai/about/connect-ai-clients/tools.md).

# Tools and source selection

After [connecting](https://docs.katara.ai/about/connect-ai-clients/connect.md), inspect your client’s `tools/list` response for available tools and input schemas. Tool discovery is not permission to execute every listed operation.

## Select the right source

For **memory and page tools**, `source_id` is an existing **Q\&A agent alias**, not a Brain ID or document ID. All require a Brain-backed Q\&A agent, except `synthesize`, which also accepts an explicitly selected collection-backed Q\&A agent. See [Q\&A](https://docs.katara.ai/about/agents-and-answering/qna.md).

Omitting `source_id`, or using literal `default`, resolves an explicitly saved Brain-backed default for the current user/client or credential within the organization. Katara does not choose the first available Brain. Save or clear a concrete Q\&A agent ID under **My Account → MCP default source**; grants have separate default-source management. An explicit alias affects that call only. Neither selection grants authority nor changes the saved default.

Two aliases for one Brain share canonical memory, not caller-private operational state. Calls do not federate multiple Brains or silently switch evidence backends.

## Seven memory verbs

| Tool           | Purpose and key inputs                                                                                                                                                                  |
| -------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `recall`       | Read active facts and optional query evidence without generation. Optional `query`; `evidence_kind` is `brain` (default), `wiki`, or `original_sources`, narrowing only query evidence. |
| `remember`     | Write a fact. Requires `fact` and nonblank `provenance`; optional entity, TTL, visibility, and write controls follow discovery.                                                         |
| `entity`       | Retrieve a deterministic entity card; requires `name`.                                                                                                                                  |
| `synthesize`   | Generate an answer under the selected Q\&A policy; requires `question`. Slower and potentially billable; prefer recall for simple lookups.                                              |
| `forget`       | Retire a fact by `id`; optional `reason`. Does not delete uploads, pages, retained prose, or history.                                                                                   |
| `context_pack` | Assemble deterministic WORLD context; requires `entities` as a string, not an array.                                                                                                    |
| `delta`        | Retrieve current page updates using `since` or `session_id`; not a lossless fact-change or deletion stream.                                                                             |

WORLD means visible to authorized Brain readers, not Internet-public. PRIVATE writes are remotely write-only, including for their author. `context_pack` and `delta` remain WORLD-only even with `include_private`. Without `since`, a new or seven-day-idle delta session starts now with an empty result; missing entries never prove cached content remains authorized.

## Six additional extended tools

| Tool              | Contract                                                                                                                                                            |
| ----------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `get_page`        | Requires `slug`; optional `include_content` returns editable canonical Markdown.                                                                                    |
| `put_page`        | Requires `slug` and complete `content`; supports revision-checked creation/update, not unrestricted overwriting.                                                    |
| `list_agents`     | No input fields; discover accessible agents and collection policies.                                                                                                |
| `get_agent`       | Requires `agent_id`; returns authorized agent information, not secrets.                                                                                             |
| `upload_document` | Requires `agent_id`, `content_type`, and base64 `data`; optional `filename` and ingestion `source_id`. This source identifier is **not** the memory alias contract. |
| `sync_collection` | Requires explicit `collection_id`; synchronizes a collection, not a Brain refresh.                                                                                  |

Uploads are ingestion, not page/fact authoring. Read a page before editing and use its observed `expected_revision`. Preserve drafts on conflict. Retry pending writes with the original request UUID and unchanged arguments; do not silently force. `force` cannot bypass authority or protected managed content.

## Permissions and availability

Memory uses `read:memory`/`write:memory`; pages use separate `read:page`/`write:page` scopes. Current membership, Q\&A/Brain authority, and contribution rights still apply. Read-only roles cannot mutate data.

Operators select `verbs` or `extended` (the configured default). A credential bearing `mcp:verbs` narrows both discovery and dispatch. Incomplete memory wiring removes memory/page tools; optional Knowledge Engine, query, and export runtime gates are separate. See [availability](https://docs.katara.ai/about/reference/availability.md).

These are MCP/API contracts, not thirteen dashboard buttons. The UI exposes selected fact and existing-page workflows, not every protocol option or a normal New page journey. See [Facts](https://docs.katara.ai/about/brains/facts.md). Never put credentials or authority overrides into tool arguments.
