GovTube

Measured conclusion: Deep-tested. L2 outcome: tools_ok.

Search what U.S. officials said on camera, with timestamps and citations. Not a government site.

Add to your agent

Copy this into your MCP client config:

{
  "mcpServers": {
    "GovTube": {
      "url": "https://api.govtube.us/mcp",
      "transport": "streamable-http"
    }
  }
}

Measured tools (7)

• govtube_search_instances read-only

  Find spoken instances of words or a phrase across official videos and White House transcripts. Returns one small page: counts for the whole result, then up to 5 videos (newest first, or most instances first with sort="most_instances"), each with up to 3 passages (quotes; overlapping instances merged), each passage listing its instances with citation id and share_url. `notes` and `more_passages` say what else there is: use `next_cursor` for more videos, govtube_list_video_instances for the rest of one video, and detail="full" for larger pages. White House transcripts have no timing (`timing_granularity` "document", null times, a `position` and a `document_url`).

  Parameters:

  • query (required) — string — Words to find. Put an exact phrase in double quotes: "no tax on tips". Unquoted words must occur within 8 words of each other.
  • sort — string — newest (default): event date, newest first; on the same day, video sessions before official transcripts. most_instances: most instances first; check order_exact in the response.
  • cursor — next_cursor from the previous page of the same search (same query and filters).
  • detail — string — compact (default): small pages, passages only. full: the larger pages of the REST API (default 10 videos x 5), with every instance listed one by one as well as in passages.
  • source — Only one source: house_floor_clerk (U.S. House floor), senate_floor_and_committees (U.S. Senate floor and committee hearings), govinfo_dcpd (White House transcripts, no video timing), govinfo_crec (Congressional Record, House and Senate speeches as printed, no video timing) or us_gov_youtube (official U.S. government YouTube channels; see source_caveats). A source can hold nothing yet: govtube_get_coverage lists what each one holds.
  • date_to — Latest event date, YYYY-MM-DD.
  • speaker — Bioguide ID of a member of Congress, e.g. 'C001053' (Rep. Tom Cole). Matches speakers a person has confirmed. Names matched only from the on-screen caption are left out (the result counts them in unreviewed_excluded) unless include_unreviewed is true. Omit to search everyone.
  • date_from — Earliest event date, YYYY-MM-DD.
  • max_videos — Videos per page, 1 to 25. Default 5, or 10 with detail="full".
  • playable_only — boolean — true: only results with playable video; transcript-only results (no video) are left out and counted in counts.transcript_only.
  • include_unreviewed — boolean — With speaker: also return moments where that person was named from the on-screen caption and nobody has reviewed the name yet. They are labelled 'matched from on-screen name, not reviewed'; say so when you quote them.
  • max_instances_per_video — Passages per video, 1 to 20 (with detail="full", also instances). Default 3, or 5 with detail="full".

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• govtube_list_video_instances read-only

  Page through one video's matches by passage, in time order. A passage is a run of instances whose excerpts overlap, shown as one quote. `offset` and `limit` count passages; `next_offset` is null on the last page. `passage_count` and `instance_count` cover the whole video. With detail="full", `instances` also lists every instance of the page one by one.

  Parameters:

  • video_id (required) — string — video_id from a search result.
  • query (required) — string — The same query as the search.
  • limit — Passages per page, 1 to 100. Default 10, or 20 with detail="full".
  • detail — string — compact (default): small pages, passages only. full: the larger pages of the REST API (default 10 videos x 5), with every instance listed one by one as well as in passages.
  • offset — integer — Passages to skip; use next_offset from the previous page.
  • speaker — Bioguide ID of a member of Congress, e.g. 'C001053' (Rep. Tom Cole). Matches speakers a person has confirmed. Names matched only from the on-screen caption are left out (the result counts them in unreviewed_excluded) unless include_unreviewed is true. Omit to search everyone.
  • include_unreviewed — boolean — With speaker: also return moments where that person was named from the on-screen caption and nobody has reviewed the name yet. They are labelled 'matched from on-screen name, not reviewed'; say so when you quote them.

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• govtube_get_transcript_window read-only

  Return speaker turns overlapping [start_ms, end_ms] with their text, for checking context. A turn with cluster null and speaker "No speaker label" holds words no voice was found for. The words are as heard; don't attribute them to a neighbouring speaker. Not available for official transcripts without timing (timing_granularity "document"): read those at the result's document_url, or resolve the citation for the surrounding turns.

  Parameters:

  • video_id (required) — string — video_id from a search result.
  • start_ms (required) — integer — Start, in milliseconds from the start of the video, e.g. an instance's start_ms minus 30000.
  • end_ms (required) — integer — End, in milliseconds; at most 5 minutes after start_ms.

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• govtube_get_video read-only

  Metadata, speakers, playback asset and the rights and use restrictions of the video's source.

  Parameters:

  • video_id (required) — string — video_id from a search result.

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• govtube_resolve_citation read-only

  Check that a cited instance still says the quoted words after corrections, and return its context. The result has a share_url to link. gd2./gd1. ids cite words in an official transcript: the result has positions instead of times, citation_text to paste, document_highlight_url, and `certifies` saying what is (and is not) certified.

  Parameters:

  • citation_id (required) — string — An instance id (gt2.... or gd2...., or an older gt1.... or gd1....).

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• govtube_get_coverage read-only

  Totals for the whole index, video and official transcripts counted apart: video sessions, hours and words transcribed; official transcripts (text, no timing); each source; date ranges, last update, items awaiting review and known gaps. Cite this with any count. For the list of sessions use govtube_list_sessions.

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• govtube_list_sessions read-only

  Every indexed item, one page at a time: video sessions and official transcripts (White House transcripts, Congressional Record speeches, one item per speech). Newest first; on the same day, video sessions before official transcripts. Each has title, date, source, length, words, whether it plays, and how many speakers are confirmed. `total` matches govtube_get_coverage's counts for the same source.

  Parameters:

  • kind — Only one kind: "video" (video sessions, timed transcripts) or "document" (official transcripts without video timing: White House transcripts, Congressional Record speeches). Omit for both.
  • limit — integer — Sessions per page, 1 to 100.
  • cursor — next_cursor from the previous page (same filters).
  • source — Only one source: house_floor_clerk (U.S. House floor), senate_floor_and_committees (U.S. Senate floor and committee hearings), govinfo_dcpd (White House transcripts, no video timing), govinfo_crec (Congressional Record, House and Senate speeches as printed, no video timing) or us_gov_youtube (official U.S. government YouTube channels; see source_caveats). A source can hold nothing yet: govtube_get_coverage lists what each one holds.
  • date_to — Latest event date, YYYY-MM-DD.
  • date_from — Earliest event date, YYYY-MM-DD.

  Hints: read-only yes · destructive no · idempotent yes · open-world no

Publisher website

Measured by MCPYP probing. Untested fields are omitted, never estimated. Listing is always free; MCPYP never proxies calls or stores user tokens.