AI and integrations
MCP tools
Every tool the MCP server offers, with the scope it needs and its arguments. Generated from the server's source.
A key sees only the tools its scopes allow. Tools return text: JSON for structured results, Markdown for pages. Failures come back as tool results with isError set.
whoami
Identity and scopes of the current key.
Scope: none, every key gets it · read-only
No arguments.
context_get
Read a context entry by namespace and key.
Scope: knowledge:read · read-only
| Argument | Type | Required | Notes |
|---|---|---|---|
namespace | string | Yes | 1–128 characters |
key | string | Yes | 1–512 characters |
context_list
List context entries in a namespace, optionally filtered by key prefix.
Scope: knowledge:read · read-only
| Argument | Type | Required | Notes |
|---|---|---|---|
namespace | string | Yes | 1–128 characters |
prefix | string | No | up to 512 characters |
cursor | string | No |
pages_list
List knowledge base pages under a parent (default: top level).
Scope: knowledge:read · read-only
| Argument | Type | Required | Notes |
|---|---|---|---|
parentId | string | No |
page_get
Read a knowledge base page as Markdown with front matter (title, version, ...). Databases include their rows as a table; rows include their property values.
Scope: knowledge:read · read-only
| Argument | Type | Required | Notes |
|---|---|---|---|
pageId | string | Yes | matches ^[A-Za-z0-9_-]{1,64}$ |
page_search
Search knowledge base pages by title and text. Returns up to 20 matches with snippets.
Scope: knowledge:read · read-only
| Argument | Type | Required | Notes |
|---|---|---|---|
query | string | Yes | 1–200 characters; trimmed |
context_put
Create or replace a context entry. Use namespaces to group related entries.
Scope: knowledge:write
| Argument | Type | Required | Notes |
|---|---|---|---|
namespace | string | Yes | 1–128 characters |
key | string | Yes | 1–512 characters |
value | string | Yes | up to 100,000 characters |
tags | string[] | No | up to 20 items; each up to 64 characters |
ttlSeconds | integer | No | > 0 |
context_delete
Delete a context entry.
Scope: knowledge:write · destructive
| Argument | Type | Required | Notes |
|---|---|---|---|
namespace | string | Yes | 1–128 characters |
key | string | Yes | 1–512 characters |
page_upsert
Create a knowledge base page from Markdown, or replace an existing page's content. Pass pageId to replace (title/icon/parentId change only when given); omit it to create a page under parentId (default: top level). For a new page without a title, a leading # Heading becomes the title. Pass baseVersion from page_get to fail instead of overwriting newer edits.
Scope: knowledge:write
| Argument | Type | Required | Notes |
|---|---|---|---|
pageId | string | No | matches ^[A-Za-z0-9_-]{1,64}$ |
parentId | string | No | matches ^[A-Za-z0-9_-]{1,64}$ |
title | string | No | up to 200 characters |
icon | string | No | up to 16 characters |
markdown | string | Yes | up to 990,000 characters |
baseVersion | integer | No | ≥ 0 |
agent_job_create
Queue work for one of the organization's self-hosted agents. http.batch runs HTTP requests limited to hosts (obeys robots.txt); exec runs a shell command (agents with full network access, or agents that limit shell commands to their allowlist); browser.batch loads up to 100 pages one at a time in the agent's installed Chrome (input { pages: [{ id, url, waitForSelector?, waitMs?, scroll?, capture?: { urlPattern, max? }, extract?: ["nextData" | "nuxtData" | "jsonLd" | "html"], select?: { nextData?: path | path[], nuxtData?: ..., jsonLd?: ... } }], delayMs? } where a path is dot-separated keys, indexes or * (e.g. "props.pageProps.products.*.name"); obeys robots.txt, identifies itself, at least 5 s between pages). ttlSeconds (60 to 604800, default 7 days) is how long the job may wait for a device. Results are delivered to callbackUrl.
Scope: agents:run · Also checks: agents:write (when args.type === "exec")
| Argument | Type | Required | Default | Notes |
|---|---|---|---|---|
type | "http.batch" | "exec" | "browser.batch" | Yes | ||
label | string | No | up to 120 characters | |
hosts | string[] | No | [] | up to 50 items; each matches ^(\*\.)?([a-z0-9](?:[a-z0-9-]{0,61}[a-z0-9])?\.)+[a-z]{2,63}$ |
input | object | Yes | values: any JSON | |
callbackUrl | string | No | URL | |
callbackSecret | string | No | up to 200 characters | |
ttlSeconds | integer | No | 60–604800 |
agent_job_cancel
Cancel an agent job: a queued job is canceled at once; a running job stops at its device's next check (status canceling, then canceled).
Scope: agents:run
| Argument | Type | Required | Notes |
|---|---|---|---|
jobId | string | Yes |
agent_job_get
Status of an agent job, with a temporary results URL once done.
Scope: agents:read · read-only
| Argument | Type | Required | Notes |
|---|---|---|---|
jobId | string | Yes |
connectors_list
Remote MCP servers connected to this organization and the tool names they add here.
Scope: connectors:read · read-only
No arguments.
Connector tools
With connectors:read, every enabled connector adds its cached tools, named <slug>__<tool>: the connector's slug, two underscores, and the tool's own name with characters other than letters, digits, _ and - replaced by _, cut at 64 characters. Their descriptions and arguments are the connector's own.
At most 200 connector tools are added per key, from the oldest connector on. If two tools end up with the same name, the first keeps it.