AirOps

Create a new AEO tag on a Brand Kit. Tags are user-defined labels that can be applied to prompts via `bulk_update_aeo_prompt_tags`. Behavior: - `name` must be unique within the Brand Kit (case-insensitive). The model enforces this via a unique index on (brand_kit_id, lower(name)). - `color` is optional. If omitted, a color is auto-assigned from the platform palette. Valid colors: light_grey, grey, green, teal, blue, purple, lilac, pink, red, coral, orange. - This tool does NOT apply the new tag to any prompts. Use `bulk_update_aeo_prompt_tags` with `operation: 'add'` to assign it afterward. IMPORTANT: Always show the user the tag you plan to create (name and color) and get explicit confirmation before calling. You can verify the tag was created by calling `list_tags` filtered by name.

Initiate a direct file upload for use with Brand Kit visual tools. This returns an upload URL (pointing directly to the storage service), upload headers, and a signed_id. The workflow is: 1. Call this tool with the Brand Kit ID and file metadata (filename, content_type, byte_size, checksum). The checksum must be the Base64-encoded MD5 digest of the file contents. 2. Upload the file to the returned `upload_url` using an HTTP PUT with **all** the provided `upload_headers`. No additional authentication is needed — the URL is a time-limited presigned URL. Example: curl -X PUT -H "Content-Type: <type>" -H "Content-MD5: <checksum>" --upload-file <path> "<upload_url>" 3. Pass the returned `signed_id` to a manage tool (e.g. manage_brand_kit_logo_variant, manage_brand_kit_visual_example, or manage_brand_kit_font). IMPORTANT: Always confirm with the user before initiating an upload.

Record a recap entry summarizing the changes you made to a Brand Kit. Call this once, near the end of a session that mutated the Brand Kit draft — not for every edit. Do not call this tool if you made no Brand Kit draft mutations this session (for example, you only read the Brand Kit, suggested edits the user rejected, or worked on Playbooks, Campaigns, or Insights). Never create a recap that says nothing changed. Put the evidence behind the changes in `body_markdown`: verbatim quotes, source URLs, and pages. Example: create_brand_kit_recap_entry( brand_kit_id: 123, title: "Refreshed tone & voice and added 2 regions", body_markdown: "Made the tone more concise per the latest brand guidelines (https://acme.com/brand). Added US East and US West to support the Q3 launch." )

Create a new empty, general-purpose grid with the given name. The grid is created with a single empty sheet (zero rows, zero columns). Use add_grid_column to add columns and write_grid to add rows. If the user belongs to more than one workspace, workspace_id is required. Do not use this tool to act on AEO insights — use create_action_grid instead, which pre-populates power step columns and rows for a given AEO action type.

Create a new sheet (grid table) within an existing grid. The sheet is created with zero rows and zero columns. Use add_grid_column to add columns and write_grid to add rows. A grid can contain at most 10 sheets.

Create a pending opportunity for a campaign. Before calling this tool, summarize the proposed opportunity name, description, and target resources for the user, then get explicit confirmation. In Quill or other OAuth MCP clients, provide play_id from list_campaigns or get_campaign. In playbook sessions, play_id is optional and is derived from the current session.

Add a web page to a Brand Kit's AEO pages. The URL is normalized before the page is created, and the page is associated with the Brand Kit's configured AEO domain. The URL must be unique within the Brand Kit. IMPORTANT: Always show the user the URL and Brand Kit you plan to use and get explicit confirmation before calling this tool. You can verify the page was created by calling `list_pages` filtered by URL.

Create a new AEO topic on a Brand Kit. Topics are categories used to group AEO prompts. Behavior: - `name` must be unique within the Brand Kit. - `color` is optional. If omitted, a color is auto-assigned from the platform palette. Valid colors: light_grey, grey, green, teal, blue, purple, lilac, pink, red, coral, orange. - This tool does NOT create or assign prompts. Use `create_aeo_prompt` with the returned topic ID to add prompts to the topic. IMPORTANT: Always show the user the topic you plan to create (name and color) and get explicit confirmation before calling. You can verify the topic was created by calling `list_topics` filtered by name.

Delete an AEO prompt from a Brand Kit. Use `list_aeo_prompts` to find the prompt ID and verify the prompt text before deletion. IMPORTANT: This action is destructive. Always show the user the exact prompt text and get explicit confirmation before calling this tool.

Delete an AEO tag from a Brand Kit. Behavior: - This is a HARD delete. The tag is removed from the Brand Kit entirely. - All taggings on prompts that referenced this tag are also deleted (cascade via `Aeo::Tag has_many :taggings, dependent: :destroy`). Every prompt that had this tag will lose it. - The response includes `tagged_prompts_count`: the number of prompts that lost the tag. Use this to communicate the blast radius back to the user. IMPORTANT: This action is destructive and cannot be undone via this tool. Always show the user the tag name AND `tagged_prompts_count` (look it up first via `list_tags` + `list_aeo_prompts` if needed) and get explicit confirmation before calling.

Delete one or more writing rules from a Brand Kit. This edits the Brand Kit draft version only; it does not change the active (live) version. A failure deleting one rule does not block or roll back the others: the response reports which rules were deleted and which could not be. IMPORTANT: Always show the user exactly which writing rules will be deleted and ask for confirmation before calling this tool.

Delete an AEO topic from a Brand Kit. Behavior: - This is a HARD delete. The topic is removed from the Brand Kit entirely. - Deletion is blocked when the topic has associated prompts. Use `list_aeo_prompts` filtered by `topic_id` to inspect prompts before deleting. - Candidate questions and question recommendations for the topic are deleted by model associations when the topic is deleted. IMPORTANT: This action is destructive and cannot be undone via this tool. Always show the user the topic name and associated prompt count, then get explicit confirmation before calling.

Discard the current prompt-assignment draft for a Brand Kit. Throws away ALL uncommitted edits — both your own and any unsaved edits the human user made in the UI — and re-mirrors a fresh empty draft from live. Live assignments are never touched. IMPORTANT: - This action is destructive and unrecoverable. Pending UI edits the user has not committed will be lost. - Always get explicit user confirmation before calling. - Calling discard when no draft exists is a no-op; a fresh empty draft is still created so further `update_aeo_prompt_assignments` calls work without setup.

Get prompts citing a specific URL. The 'id' parameter is the URL to look up.

Get a specific page content update by ID. Its change_detection object carries the detection status and categories, plus a nested changes array (the before/after diff) on this single-update fetch. Track content updates.

Inspect the current prompt-assignment draft state for a Brand Kit without modifying anything. Read-only. This is the authoritative source for workspace estimated-answers numbers (live and draft). Call it whenever you need them — never compute or guess them yourself. In particular: - At the start of a session, to detect a pre-existing draft from unsaved UI edits. - Immediately after `commit_aeo_prompt_assignments`, to read the new live `estimated_answers` — the commit tool does not return them. - Before staging more changes via `update_aeo_prompt_assignments`, to compare against `max_answers`. Returns: - `has_draft`: whether a draft session exists for this brand kit. - `has_changes`: whether the draft has uncommitted edits relative to live (null if no draft). - `draft_id`: the draft's ID (null if no draft). - `live_estimated_answers`: current workspace estimated answers/month based on live assignments. - `draft_estimated_answers`: what the workspace estimated answers/month would be if the draft were committed now (null if no draft). - `max_answers`: the workspace's effective answers limit. - `limit_exceeded`: whether committing the draft now would exceed the limit (false if no draft).

Get a specific AI answer by ID with full text content. Get a single AI answer with full details. Use this to retrieve the complete answer text (not truncated), citations, and mentions. Use includes parameter to get citations, mentions, and persona details.

Fetch a Brand Kit's brand identity (writing_tone, writing_persona) and associated entities (product lines, audiences, content types, regions, writing rules, custom variables, visual/design assets). Call AFTER `list_brand_kits` confirms `brand_management_enabled: true`. # Decision flow 1. Specific entity by ID → `includes=["<type>"]`, `filters=[{field: "<type>.id", operator: "EQ", value: <id>}]`. 2. Entity by name (product_lines, content_types, audiences, regions, custom_variables, visual_examples) → `filters=[{field: "<type>.name", operator: "CONTAINS", value: "<fragment>"}]`. 3. Find text content ("rule about tone", "any mention of Red List") → filter by the text field: `filters=[{field: "writing_rules.text", operator: "CONTAINS", value: "tone"}]`. 4. Structural discovery → small `includes` with `fields=["<type>.id","<type>.name"]` only. # Scale (typical kit vs outliers) Most kits are small — these rarely need filtering: - audiences, regions, usage_rules, palettes, content_samples: usually 0-2, rarely >10 - content_types, product_lines: usually 1-3 These have tails you must handle: - writing_rules: usually 3, outliers exceed 200. Never include in full without a filter — filter by text (CONTAINS) or by id. - competitors: usually 10, outliers exceed 250. Include via product_lines.competitors only when needed, and consider a name filter. - custom_variables: outliers exceed 150. Filter by name (CONTAINS) for known variables (e.g. "Red List"), or by value (CONTAINS) for content. - Scoped writing_rules via audience/region/content_type nesting: outliers exceed 200. Filter the parent by id or name first. # Writing rules — global vs scoped Writing rules are EITHER global OR scoped to a single audience/content_type/region (never both). For content targeting a specific audience/region/content_type, fetch BOTH: 1. Top-level `writing_rules` include → global rules. 2. Nested writing_rules on the scoping parent → scoped rules. Merge client-side. # Anti-patterns - `includes=["writing_rules","custom_variables","product_lines.competitors", "content_types.writing_rules"]` without filters → blows context on outlier kits. - Calling without `fields` on a large kit → returns full text (template_outline, rule text, ICP) for every row. - Forgetting filters on `writing_rules`, `custom_variables`, or `competitors` — those are the three biggest tails. # Versioning `version: "draft"` reads pending edits; defaults to `"active"`. Use active for on-brand output unless the user explicitly asks for draft.

Get a campaign by ID, including action grid IDs needed to inspect or update its grid. Campaigns are Plays that coordinate strategy playbooks, action grids, and related opportunities for improving AEO performance.

Check the status of grid row executions. Returns the overall status and per-column detail for each execution. Use this after run_grid_rows() to poll for completion. Maximum 50 executions per call.

Get AEO insights configuration for a Brand Kit, this includes the relevant information to use any AEO and analytics tools. Returns competitors, personas, prompt count, and other relevant information. Use `list_brand_kits` first to find Brand Kits with `aeo_enabled: true`.

Get AEO metrics for a specific web page. Page details include citation share, citation rate, unique cited questions count, and Google Search Console metrics (clicks, impressions, position, CTR) with trends.

Get prompts citing a specific web page. Returns AI prompts that cite the page along with citation metrics (citation_rate, mention_rate) and trends. Use includes: ["topic"] to get topic details for each prompt.Prompts are the questions that can be asked about a brand or web page. They are the search queries that are used to get AI answers with citations and mentions.

Get AI answers for a specific prompt/question. Prompt answers are the AI answers for a specific question/prompt asked to multiple AI providers and the answers are aggregated. The answers include whether the brand was mentioned and cited in the answer. To get full answer text, citations, and mentions, use the get_answer tool with the answer ID.

Get a specific report by ID with its module configurations. Reports are saved analytics views for a Brand Kit. They contain modules that display different analytics metrics.

Get individual AI answers with sentiment details for a specific theme. Returns answer text, sentiment (positive/neutral/negative), confidence score, and provider. Use this to understand WHY a theme has a particular sentiment score. Today's data is still being processed — always use yesterday (2026-08-20) or earlier as end_date.

Step 2 of the two-step file ingestion flow for a Knowledge Base. Consumes a `signed_id` returned by `knowledge_base_create_direct_upload` (step 1) plus the file's metadata, and registers the document with the Knowledge Base. Returns immediately with the new `document_id` in `pending` status — poll `knowledge_base_get_status` to monitor indexing. Workflow: 1. Call `knowledge_base_create_direct_upload` with the file metadata to get back a `signed_id` and a presigned `upload_url`. 2. PUT the file bytes to the `upload_url` with the provided headers. 3. Call this tool with the `signed_id`, the desired `name` (used as the document's display name), the `document_type` MIME (must match the supported types), and an optional `metadata` hash. The `metadata` hash, if provided, is a single-level key/value object. It **replaces** any existing metadata on the document (no merge) and is filterable at search time via `search_knowledge_base`.

Bulk-ingest one or more web pages into a Knowledge Base. Each URL becomes a separate document that fetches and indexes asynchronously. The call returns immediately with the new document IDs in `pending` state — poll `knowledge_base_get_status` to check progress. URLs must be absolute and include the `http://` or `https://` scheme (e.g. `https://example.com/page`). No documents are created if any URL fails validation. The `metadata` hash, if provided, is applied identically to every document in the batch (single-level keys only); it replaces (not merges) any existing metadata and is filterable at search time.

Initiate a direct file upload for a Knowledge Base. Returns a presigned S3 upload URL, the required upload headers, and a `signed_id` you'll use with `knowledge_base_add_file` to register the document. This is the first call in the two-step file ingestion flow — large files (PDFs, DOCX, MD, HTML, etc.) don't fit through the MCP transport, so the file bytes go directly from your client to S3. Workflow: 1. Call this tool with the file metadata (filename, content_type, byte_size, checksum). The checksum must be the Base64-encoded MD5 digest of the file contents. 2. PUT the raw file bytes to the returned `upload_url` with **all** the provided `upload_headers`. No additional authentication is needed — the URL is a time-limited presigned URL. 3. Pass the returned `signed_id` to `knowledge_base_add_file` to register the document. Only after step 3 is the file searchable. Example: > knowledge_base_create_direct_upload(knowledge_base_id: 1, filename: "report.pdf", content_type: "application/pdf", byte_size: 1234567, checksum: "...") # → { signed_id: "abc...", upload_url: "https://s3...", upload_headers: { ... } } > knowledge_base_add_file(knowledge_base_id: 1, signed_id: "abc...", name: "Q4 Report", document_type: "application/pdf") # → { document_ids: [42], status: "pending" } Maximum file size is 256 MB.

Permanently delete a Knowledge Base and ALL of its documents. This cascades through every document in the KB and drops the underlying vectors. This action cannot be undone. IMPORTANT: Always warn the user that deletion is permanent and irreversible, name the Knowledge Base being deleted, and ask for explicit confirmation before calling this tool.

Permanently delete a single document from a Knowledge Base. This action cannot be undone. IMPORTANT: Always warn the user that deletion is permanent and ask for explicit confirmation before calling this tool.

Read the full reconstructed text content of a Knowledge Base document end-to-end — the loader-extracted text from every chunk concatenated in `position` order. Use when chunked search results aren't enough: summarizing a whole document, answering questions across an entire report, or reading back a doc before delete-and-recreate. For finding specific passages, prefer `search_knowledge_base`. Returns up to `max_chars` characters starting at `offset` (defaults to 0 and 200000). When the document is larger than the slice, the response includes `next_offset` (the value to pass on the next call to continue reading); when fully read, `next_offset` is `null`. For documents still indexing (`status: pending`) or in error (`status: error`), content may be empty — check `status` first. Content is reconstructed by joining chunks; minor overlap or duplication between adjacent chunks is possible depending on the original loader's chunking strategy. For the original source file, use the existing download path outside MCP.

Get the indexing status of a Knowledge Base and its documents. Returns a Knowledge Base–level rollup (status, pending and total document counts) plus a paginated list of per-document statuses. Use this to monitor indexing after writes — only documents with status "ready" are returned by search_knowledge_base. Pass the returned `cursor` back to fetch the next page; pass `document_ids` to filter to specific documents.

Create or update a Knowledge Base. Omit `knowledge_base_id` to create a new one; pass it to update an existing one. On create, `name` is required; pass `workspace_id` if you have access to more than one workspace. On update, only the fields you pass change.

Replace a document's user-facing metadata in full. Accepts a single-level hash that **replaces** (not merges) the existing user-facing metadata. To remove a key, pass the full new hash that omits it. To clear all metadata, pass `{}`. Filterable at search time via `search_knowledge_base`. Metadata-only — does not re-embed the document. Loader-derived metadata (scraped page metadata for URLs, chunker metadata for files) is not touched. Search filters may take a few seconds to reflect the new values. The document must be fully indexed (`status: ready`) before calling this tool. If indexing isn't complete, the tool returns a validation error.

List citations (URLs) with metrics for a Brand Kit. Citations (URLs) with detailed metrics including influence score. Shows individual URLs being cited in AI answers with page-level context.

List domains cited in AI answers for a Brand Kit. Cited domains aggregated by domain with citation metrics. Shows which domains are being cited in AI answers for a Brand Kit.

List page content updates for a workspace. Track content updates.

List AEO prompts for a specific Brand Kit. Questions are the AI prompts that can be asked about a brand. They are used to get AI answers with citations and mentions.

List AI answers for a brand kit with filters for date range, providers, countries, prompt_id, and brand_mentioned. Individual AI answers with their cited URLs and brand/competitor mentions.