Reading Know from MCP and the CLI

Everything in Know can be read by an agent over the Operator MCP, or from a terminal with the leed CLI. Those are the only two routes Know data leaves by. Both use your own sign-in and your role, so they return what you can see in Know and nothing more. There are no CSV exports, no webhooks and no emailed reports.

The Data access screen

Data access: MCP & CLI (/know/data-access), the link at the foot of the Know sub-nav, shows both routes:

  • Operator MCP: the server’s address with a Copy button, OAuth sign-in on first use, and the tools that read Know with what each returns: get_know_view, run_saved_view, save_view, get_analytics and list_analytics_metrics.
  • leed CLI: the leed know commands, every one of which takes --json. Rows that name a person follow leed contacts export: they are written with --out to a file only you can read (mode 0600), never printed.
  • How a saved view gets made: from any view’s Save view, from an answer in Ask, or by an agent with save_view.
  • Recent access: Operator MCP reads in the last 30 days, newest first, by person, with each call’s outcome (succeeded, failed, declined or pending). It needs agent analytics (Growth). Anonymous agents reading your docs are not listed here; they are counted on the Agents view. Reads made with leed know are not recorded.

Over MCP

get_know_view with no view lists every view: the filters it takes, the sections it returns, the sections that name people, and the plan feature it needs. With a view, pass start and end (ISO 8601, at most 365 days) and any of that view’s filters. The window is clamped to your plan’s analytics history, and the clamped window comes back with the result.

  • A filter the view does not take is refused, with the view’s list. So are combinations that mean nothing, such as an agent filter with readers: "people", or an account with anonymous agents only.
  • A view your plan does not include is still a success: gated: true, its totals, and no sections. Inside a view every plan sees, a gated section returns its real total in place of its rows.
  • Filters a view lists under peopleOnlyFilters (referrer and channel, on Overview and Pages) narrow people’s figures only, since agent reads have no referrer or channel.
  • Ids carry from one view to the next: a contactId from People opens the person view, an accountId the account view, a pageId the page view, and a release name the release view.

run_saved_view runs a saved view by id, or lists the saved views you can see; save_view saves one. The Search & gaps tools are get_gap, mark_gap_in_progress, dismiss_gap (with an optional reason) and reopen_gap. get_analytics reads a single metric. Every tool is listed on MCP tools: analytics and introspection.

From the CLI

leed know views lists the views, their filters and your saved views. leed know show <view> reads one, with the same filters as flags (--readers, --agent-identity, --country, --account and the rest) and --filter key=value for a view’s own filters. The window is the last 30 days unless you pass --start and --end; Campaigns defaults to the selected campaign’s dates, as on the screen. leed know run runs a saved view.

Sections that name people never go to standard output. Pass --out <file> and they are written there as JSON with mode 0600; a file inside a git repository that does not ignore it is refused. Every command and flag is on leed know commands.

Rules both routes share

  • Agents are signed in or anonymous. There is no breakdown by AI client, and none is estimated.
  • People and accounts are counted only inside the number of contacts your plan lets you see.
  • Nothing names a person except the sections that name people, and those go only to a file from the CLI.

The leed:analytics plugin skill answers questions about how your content performs through these same tools. Saving and asking are covered on saved views and Ask.

ESC