USLege

Create a track: either a client the user represents (trackType 'Client') or a subject they watch (trackType 'Policy' or 'Other'). A track is a named collection of bills that the user wants to follow. Call list-tracks first and check for a track of the same name, so running a setup conversation twice does not create every client again. Status, size and the deal fields describe a client and should be left out for a policy area. Set them here rather than calling update-track afterwards, so a client is never left half-created. When trackAccessType is 'Group', provide groupMemberEmails with the email addresses of users to share with. Emails will be resolved to user IDs automatically. Before creating, confirm the track name and settings with the user.

create-track

Create a new portal owned by the calling user. Caller must have access to every track id provided; the call is rejected if any track is inaccessible. New portals are created with `accessType: 'Owner'` (only the creator can see them) — call `share-portal` afterward to expand access to a Group or to the whole organization. Asset uploads (logos, banners, attachments) are NOT supported via MCP — direct users to the UI for that step. Before creating, always confirm the portal name, track list, and visibility flags with the user.

create-portal

Create a new video radar rule owned by the calling user. Video radar rules are saved natural-language queries that monitor legislative video transcripts for matches. They are persisted as `AiMonitorQueries` rows (with `AiMonitorQueriesByInstitution` for targeting and `AiMonitorSharedUsers` for the recipient list) — there is no separate "RadarVideoRule" table. Targeting accepts either individual institution ids (`organizationIds`), state-level "all institutions of type X" selections (`allSelections`), or both — at least one must be provided. State / institution access is enforced by the existing tRPC guard; unauthorized targets are rejected. `alertRecipients` is required: `Owner` (only the creator), `CustomList` (resolves `selectedEmails` against the caller's org), or `OwnerOrganization` (every user in the caller's org). Before creating, always confirm the rule definition, targeted institutions / "all" selections, alert recipients, and notification cadence with the user.

create-video-radar-rule

Create a new organization-wide tag, visible to everyone in the user's organization. Call list-organization-tags first. If an Organization-access tag with the same name already exists, use it instead of creating a new one. If the same-named tag is Owner or Group access, it is not visible to the whole organization: tell the user, and ask whether to use it as is or create an org-wide tag under a different name (names are unique per organization). If creation fails because the name is taken but the tag is not in list-organization-tags, it is a colleague's private tag: ask the user for a different name. The returned tag ID can be passed straight to add-bill-tag as organizationTagId. Before creating, confirm the tag name, color and description with the user.

create-organization-tag

Create or update a video clip from a legislative video recording. **Create** (omit clipId): Creates a new clip from the specified video. Requires videoId, title, startTimeSeconds, and endTimeSeconds. By default, the clip is saved but NOT processed for download — set processClip to true to trigger processing. **Update** (provide clipId): Updates an existing clip you own. Only provided fields are changed. If start/end times change and processClip is true, clip processing is re-triggered. Do NOT process the clip for download unless the user explicitly asks. Processing creates a downloadable video file from the source video — it takes a few minutes and costs resources. Leave processClip unset (defaults to false) unless the user says they want it processed. Logos default to none. Before creating a clip, ask the user if they want a logo watermark applied. If yes, call list-logos first to show available options and let them choose. Pass the chosen logoId and optional logoPlacement. Before creating or updating, confirm the clip details with the user — especially the video, timestamps, and title. Requires the video ID. Call search-videos first to find the video. Use get-video-transcript or search-transcript to identify the right start/end timestamps; capture all the relevant discussion the user asked for and err on the side of slightly too long over too short. After creating a clip, you can provide the user a shareable link: https://app.uslege.ai/share/{clipId}?from=clips

manage-clip

Delete a comment you previously posted on a bill, along with any of its attachments. Only the comment's own author can delete it — call list-bill-comments first to find the comment ID and confirm isMine is true. This action cannot be undone. Before deleting, always confirm with the user by showing the comment's content.

delete-bill-comment

Permanently delete a bill radar rule. Deletion cascades to the rule's stored bill matches and notification records via the underlying foreign-key constraints — there is no undo. Before deleting, always confirm with the user by showing the rule name and any pending shared users.

delete-bill-radar-rule

Delete a calendar alert rule. Only the rule owner may delete. Removes the rule, its location associations, and every shared-user row. Confirm with the user before calling — this is irreversible.

delete-calendar-alert-rule

Delete a dashboard. Owner-only. Cascades to every widget on the dashboard, every member entry, and every per-user view order — all permanently lost. Before deleting, ALWAYS confirm with the user. Call get-dashboard first and show them the dashboard name, widget count, and access type so they can identify what's about to disappear. The operation cannot be undone.

delete-dashboard

Delete a single widget from a dashboard. Permanent and cannot be undone. Before deleting, ALWAYS confirm with the user. Show the widget name and type (call get-dashboard to retrieve them if you don't already have them). Caller must be Owner or Editor on the dashboard; Viewers are rejected. After deleting, consider calling reorder-widgets to backfill the gap if the user wants a tidy grid — let the user choose; some prefer the gap as a placeholder.

delete-widget

Delete a bill track and remove all associated data. This permanently removes the track along with its tracked bills, digests, video clips, and access records. The user must have access to the track to delete it. This action cannot be undone. Requires the track ID. Call listTracks first to find the ID of the track to delete. Before deleting, always confirm with the user by showing the track name and bill count.

delete-track

Permanently delete a portal. Owner-only — non-owner callers receive a permission error. Cascades cleanup of: portal-track associations, portal access entries, free email-access entries, portal-level updates, and uploaded assets (logos / banners / attachments in Supabase storage). This is irreversible. Underlying tracked bills and source tracks are NOT deleted — only the portal wrapper. Before deleting, always confirm with the user by showing the portal name and (if possible) the track count + recent update count obtained via `get-portal`.

delete-portal

Permanently delete a portal-level update. Cleans up any file attachments stored in Supabase, then deletes the update row. Irreversible. Requires paid access to the parent portal. Before deleting, confirm with the user — show the update title (or first line of content) and `updateDate` from `list-portal-updates`.

delete-portal-update

Permanently delete a video radar rule. Video radar rules are persisted as `AiMonitorQueries` rows — there is no separate "RadarVideoRule" table. Deletion is restricted to the rule's owner; sharing with the caller is not enough. Associated `AiMonitorQueriesByInstitution` rows are removed, while historical `AiMonitorMatches` rows are kept for audit. This action cannot be undone. Always confirm the rule id and intent with the user before calling.

delete-video-radar-rule

Delete a client from the calling user's organization. This cannot be undone. Requires the client id — call list-clients first to find it. Before deleting, always confirm with the user by showing the client's name.

delete-client

Deprovision a Clerk organization: remove EVERY member while keeping the org and all of its data (tracks, dashboards, settings, Salesforce/JustPaid links, states) so a returning client can be reactivated later. Super-admin tool. This is the MANUAL path for churned or overdue paying customers — trials deprovision automatically when they expire, so this tool is normally not needed for them. Two-step flow, and the first step is mandatory: 1. Call with just the org id (no confirmationToken). NO changes are made — the tool returns status "confirmation_required" with the org name, every member who will lose access (names/emails/roles), trial + billing status, the org's Salesforce/JustPaid links, and a confirmationToken. Show ALL of this to the operator and get their explicit approval. 2. Only after the operator approves, call again with the same org id plus the returned confirmationToken to execute. If they decline, make no further calls. The token is bound to everything the preview showed — the org's membership set plus its trial/billing status and Salesforce/JustPaid links. If any of that changes between the two calls the token stops matching and the tool errors — re-run step 1 and re-review. Never invent or guess a token. Requires the Clerk org id (org_...). Call search-orgs first to find it. Already-deprovisioned orgs error clearly instead of no-opping. Removal is resumable: if a member removal fails partway, the error reports progress and re-running step 1 then step 2 removes the rest (the deprovision stamp is only written once every member is removed). The removed roster is captured first, so this is reversible: reactivate-organization re-seats everyone and clears the stamp in one call.

deprovision-organization

Duplicate a track into a new track the user owns, with the same type, description, sharing, bills, laws and (when the user can track them) interim topics. Call list-tracks first to find the source track ID. Before duplicating, ask the user about each optional part to copy rather than guessing, then confirm the new track's name and those choices with the user before calling. Bills from states the user is not entitled to are left out. Copied comments are new comments that keep their original authors; mentions are not copied and nobody is notified. Portal, client, radar rules, priorities, tasks and notification settings are never copied.

duplicate-track

Modify an existing portal-level update (announcement). Partial-field semantics — at least one changeable field must be provided; omitted fields are unchanged. Attachments are managed in the UI — this tool does not add or remove file attachments. Before updating, confirm the changes with the user.

update-portal-update

Exports a video's full transcript as a downloadable file. Defaults to a DOCX file. Pass format: "json" or "md" only when the user explicitly asks for that file format — otherwise leave it unset so the user receives a DOCX. When format is "json", the downloaded file is UTF-8 JSON with this shape: { "videoId": number, "title": string | null, "institutionName": string, "recordedOn": "YYYY-MM-DD", "paragraphs": [{ "startSeconds": number, "endSeconds": number, "text": string }] } When format is "md", the downloaded file is a UTF-8 Markdown document with an institution-name heading, a title heading, a recorded-on line, and one paragraph of transcript text per line — no timestamps or speaker labels. Returns a presigned download URL (valid for 1 hour) along with video metadata and a suggested filename. The AI client should surface the downloadUrl to the user as a link or download action — there is no inline file content to decode. Requires a video ID. Use the search-videos tool first to find the video ID. To read the transcript text directly (e.g., to summarize or quote it), use get-video-transcript instead — that tool returns the paragraphs inline.

export-transcript

Fetch the full file body (HTML or text) for state register sections — the same content the in-app RegisterViewer shows when you click a section in the tree. Get sectionIds from get-state-register-issue-inventory (whole issue) or search-state-register-sections (keyword). Two modes: - sectionId (optionally fileId): one section, full body, top-level fields populated. - sectionIds (up to 10): several sections in one call, returned in `items` in the order requested, each with its default (first) file. Use this when summarizing an agency's or an edition's rules — ten rules is one call, not ten. Bodies are capped at 15000 characters each by default (truncated=true when cut); raise maxCharsPerSection if you need more of a specific rule, or re-fetch it alone. Any per-section failure (missing section, no files, state not authorized, or a transient storage/database fault reported as error="fetch_failed") is reported inline on that item and does not fail the others; retry just the failed sectionIds. A section may have one or more attached files. Every item includes a `files` array listing every file's id/name/type, so a single call gives you full visibility into what's attached. For TEXT-LIKE files (HTML/HTM/TXT/MD/JSON), the item includes the decoded body as `content`. For BINARY files (PDF/DOC/DOCX), `content` is null and the item instead includes `downloadUrl` — a presigned S3 URL (valid 1 hour, with `expiresAt` ISO timestamp) you can surface to the user as a download link or fetch yourself to parse the file contents. By default `selectedFile` is the first attached file. To read a different file, call again in single mode with that file's id from the `files` array as the `fileId` input. Errors (top-level in single mode, per item in batch mode): - error="section_not_found" if a sectionId doesn't exist - error="no_files_found" if the section has no attached files - error="file_not_attached" if fileId is provided but isn't attached to this section - error="state_access_denied" if your org isn't authorized for the section's state - error="fetch_failed" (batch mode only) if that one section could not be read right now

get-state-register-section-content

The mandatory duplicate check for a failed video: returns other videos sharing the SAME institution + committee + recordedOn as the target. Use this before recommending deletion — a target may only be marked for deletion when a fully-processed duplicate exists AND the target itself has no usable content. Get the videoId from dev-videos-triage-failed. `isFullyProcessed` flags a candidate whose latest jobs include a Completed PostprocessTranscription (a real, processed duplicate). Different durations/rooms usually mean separate recordings, not duplicates — verify before concluding. A same-day sibling from a different `sourceViewUrl` may be a true duplicate (a livestream archive superseded by a full re-upload), one segment of a meeting the extractor split across uploads (a livestream that dropped and resumed, an executive-session return), or a different meeting the same body held that day. Only call it a duplicate with positive evidence of the same recording: the same meeting per the titles, the same start time per the two source pages, and `seconds` covering the target's source-reported duration. A `part` on either side, or a candidate far shorter than the target's source duration, marks a split meeting whose failed segment is the only copy of that part of the meeting — retry or defer it, never delete it. When the evidence is incomplete, defer rather than delete. Errors if the video does not exist.

dev-videos-triage-duplicate-candidates

Get per-year lobbyist <-> client compensation rows for Texas. Filterable by lobbyistId, clientId, year, and compensationMethod (Earned / Paid / Prospect). Sorted by year desc, then lobbyist name asc. Each row returns the lobbyist + client refs (with names), reporting interval, compensation method, minAmount/maxAmount in USD, the original compensationCode, and compensatedFrom/compensatedTo windows (ISO 8601). Lobbyist data is currently TX-only — non-TX-authorized callers receive a structured access-denied error.

get-lobbyist-compensation

Get a Texas lobbyist's full profile by id. Returns: - contact info (full name, firm, address, phone) - subject matters lobbied on, with the year each was reported - per-client compensation history (minAmount/maxAmount in USD, compensationCode original code, method, reporting interval, compensated-from/to dates) - political fund reimbursements (fund name + address + reporting interval per year) Lobbyist data is currently TX-only — non-TX-authorized callers receive a structured access-denied error. Returns { found: false } when no lobbyist exists for the supplied id.

get-lobbyist-details

Fetch the organization-level notification settings for a single bill and (when applicable) its current subscriber list. Call `list-bill-subscriptions` or `search-bills` first to discover the bill id. Returns `settings: null` when the caller's organization has no notification settings configured for the bill. The `subscribers` array is only populated when the bill's access type is `Group` — for `Owner` / `Track` / `Organization` access types it is empty by design, since per-user subscriptions are not tracked on those rows.

get-bill-notification-settings

Fetch a single calendar alert rule by id. Call `list-calendar-alert-rules` first to discover the rule id. Returns `rule: null` when the rule does not exist or the caller has no access — the two cases are intentionally indistinguishable.

get-calendar-alert-rule

Consolidated profile for a single campaign-finance donor (entity), assembled in one call: identity header, lifetime giving stats, the top recipients they've funded, and per-section row counts. This is the donor detail view. First call search-campaign-finance to resolve a name to an entity id (results where kind='entity'); pass that id here. If the donor is itself a registered filer, the header's resolvedFilerId points to the filer-side record — use the filer tools for that view. Money amounts are integer cents. `topRecipients.rollup` buckets ALL of the donor's giving by recipient type (candidates / committees / party / other), while `topRecipients.recipients` is only the five largest recipients by dollars. `sectionCounts` reports how many rows exist per detail section so you can tell what's worth drilling into. Campaign finance is only available for states where the "elections" product is enabled. A state your org isn't authorized for returns `state_access_denied`; an authorized state without campaign-finance data returns `state_not_supported` along with the currently-supported states. An unknown entity id (or one from another state) returns `not_found`.

get-entity

One page of a single donor's (entity's) transaction history for one section, the drill-down behind get-entity's `sectionCounts`. Resolve a donor id with search-campaign-finance (results where kind='entity'), call get-entity to see which sections have rows, then page the section you want here. Each row's `counterparty` is the filer on the other side; pass its `filerId` to the filer tools. Rows are newest-first by transaction date (id-tiebroken). Paginate with `pagination.offset`: add `limit` to `offset` for the next page and stop once `offset + rows.length >= totalCount`. An unknown or wrong-state donor id simply returns zero rows (never an error). Campaign finance is only available for states where the "elections" product is enabled. A state your org isn't authorized for returns `state_access_denied`; an authorized state without campaign-finance data returns `state_not_supported` with the currently-supported states.

get-entity-activity

One paginated activity section for a single campaign-finance filer, selected by `type`. Use it after get-filer once you have a filerId and its `sectionCounts` tell you which sections have rows — this is the drill-down that returns the actual transaction rows behind those counts. Discover the filerId first with search-campaign-finance (a result where kind='filer', or the resolvedFilerId on a donor). Call once per section you want; the response populates only the field named by `type` (e.g. type='contributions' fills `contributions`), and echoes `type` back. Pagination: `contributions`, `expenditures`, `reports`, `pledges`, `loans`, `spacPositions`, and `travel` are server-paginated — use `limit`/`offset` and page against `totalCount`. `topContributors` and `spendByCategory` are aggregates that ignore `limit`/`offset`: `topContributors` returns only the top 5 donors by total (a fixed cap — `totalCount` is at most 5 and does NOT reflect the true number of distinct contributors), while `spendByCategory` returns every category and its `totalCount` is the full category count. Period bounds: `dateFrom`/`dateTo` scope only the money sections `contributions`, `expenditures`, `topContributors`, and `spendByCategory` (on transaction date). The all-time sections `reports`, `pledges`, `loans`, `spacPositions`, and `travel` ignore them. Campaign finance is only available for states where the "elections" product is enabled. A state your org isn't authorized for returns `state_access_denied`; an authorized state without campaign-finance data returns `state_not_supported` along with the currently-supported states.

get-filer-activity

Get a committee's calendar of past and upcoming meetings/hearings, including agenda bills, participants, hearing notice links, and attached files. Fields not present in the underlying data (room, structured participants, isHost) are returned as null/empty. If a state does not have the "calendars" product enabled, returns an empty events list with an explanatory message rather than an error. If the committee does not exist OR the user is not authorized for the committee's state, returns { events: [], totalCount: 0, error: "not_found" } with the same shape — these cases are intentionally indistinguishable to prevent committee-ID enumeration.

get-committee-calendar

Get a committee's full detail plus its seated members, staff roster, parent committee, and direct subcommittees (one level deep only). Returns ONLY this committee's own membership (chair, vice chair, ranking minority chair, members of this committee). It does NOT return the full chamber/legislature roster — for that, use list-legislators. For per-meeting attendance, use a calendar event tool instead. Legislators are sorted Chair → ViceChair → RankingMinorityChair → Member, then alphabetical. Staff is sorted alphabetical by name; a staffer with multiple roles surfaces once with all StafferPositionType values aggregated into positions[]. staff[] is the committee's own professional staff. For the members' personal-office staff (e.g. each member's Legislative Director or Chief of Staff), call list-staff with this committeeId and a title filter instead of calling list-staff per member.

get-committee

Consolidated profile for a single campaign-finance filer (candidate, officeholder, or PAC/committee) in one call: identity header, summary-card aggregates, headline money-in/money-out totals, officers, legislator links, and per-section row counts. Use this as the follow-up to search-campaign-finance once you have a filer id — it replaces six separate lookups. Discover the filer id first with search-campaign-finance (a result where kind='filer', or the resolvedFilerId on a donor). `sectionCounts` tells you which detail sections have rows so you can decide which paginated follow-up tools to call next. `moneyTotals` is the only period-aware section: pass `dateFrom` / `dateTo` to scope it, or omit both for all-time; the other sections are always all-time. `header` is null when no filer matches the id in the given state. Campaign finance is only available for states where the "elections" product is enabled. A state your org isn't authorized for returns `state_access_denied`; an authorized state without campaign-finance data returns `state_not_supported` along with the currently-supported states.

get-filer

Get one dashboard's full layout: metadata, the calling user's role on it, the member list (user id + role), and every widget on the grid with its name, type, display type, stored filters JSON, and grid position. Useful for understanding what a user is currently tracking before suggesting changes or fetching rendered data. Requires the dashboard id, which the user must provide directly. Member user ids returned here are internal Clerk user ids scoped to the caller's organization.

get-dashboard

Get a downloadable link for one document stored against a legislative calendar event — handouts, witness lists, minutes, presentations and the like. Call get-calendar-event-details first and take a fileId from its "materials" array; this tool takes no other way of naming a document. Returns a presigned download URL (valid for 1 hour, with an `expiresAt` ISO timestamp) alongside the document's type and label. Surface it to the user as a download link, or fetch it yourself to parse the contents — there is no inline file content to decode. A document in a state you cannot see reports the same error as one that does not exist.

get-calendar-event-document

Fetch a single portal by id with its tracks, paid-access users, and (if the caller is owner / paid) the email-access list. Call `list-portals` first to discover the portal id. Returns `portal: null` when the caller has no access. The `paidAccessUsers` array is empty for read-only viewers. The `emailAccess` array is populated only for callers with paid/owner access (it is owner-managed data) — for everyone else it is `null`.

get-portal

Fetch a single executive order by database ID, including the full body, scraper- supplied summary, source URL, and PDF link. Federal EOs carry state='US' with the president in the 'governor' field. Use after identifying an EO via search-executive-orders.

get-executive-order

Fetch a single press release by database ID, including the full body (HTML) and source URL. Releases with no state (federal / unknown jurisdiction) carry state=null. Use after identifying a release via search-press-releases.

get-press-release

Fetch a single social media post by its internal id, including its full text, the account that surfaced it, and the bills it references. On a Repost the text is the original author's words, not the account holder's — read `originalAuthor` to attribute the argument correctly rather than crediting the reposter. Find the id via search-social-posts. Returns "not found" for an id that doesn't exist or belongs to a state the caller cannot access; the two are indistinguishable on purpose.

get-social-post

Fetch the notification settings for a single track and the current recipients list. Call `list-tracks` or `list-track-subscriptions` first to discover the track id. Errors with "Track has no notifications configured yet" when the track exists but has no `TrackNotificationSettings` row yet. The `recipients.subscriberUserIds` array reflects the users who currently receive notifications; `recipients.accessUsers` lists every user who has access to the track (the candidate pool for `update-track-notification-settings.recipientUserIds`).

get-track-notification-settings

The spine of a single-video failure investigation: returns one video with its institution, committee, and FULL job-queue history (every attempt, error category, and error metadata), newest job first. Get the videoId from dev-videos-triage-failed. Use the latest failing job's errorCategory + errorMetadata to identify the failure, then dev-videos-triage-source-urls to fetch the URLs for a live source check. Errors if the video does not exist.

dev-videos-triage-jobs

Returns the source URLs and identity for one video — the argument feed for the sandbox source-check scripts (check-source.ts / check-source-listing.ts). Fetch sourceViewUrl (the immutable page URL) and sourceFileUrl (the resolved media URL) to verify whether the source is still live, has moved, or is a placeholder. Get the videoId from dev-videos-triage-failed or dev-videos-triage-jobs. Storage keys are intentionally not exposed — hasStoredVideo/hasStoredAudio are DB flags telling you only whether a key is recorded (not whether the S3 object is actually present); verify that with dev-videos-triage-s3-artifacts. Errors if the video does not exist.

dev-videos-triage-source-urls