Meltwater

Use this tool when the user wants to save a reusable search configuration for contacts or sources. This is a write operation that creates a new saved search visible to all workspace members. Returns the created saved search object with its ID, searchConfiguration, and metadata. Parameters: - searchName: Display name (3-200 characters) - searchType: "contacts", "sources", or "ai-contacts" - keywordType: "named" (by name), "writing" (by topics they write about), or "publishing_at" (by publication) - searchConfiguration: The search filters, query, and pagination to save - relevantConfiguration: Relevance settings including query, filters, dateStart, dateEnd Limit: 200 saved searches per company. Always confirm with the user before creating to avoid duplicates. Do NOT use this tool to execute a search — use mediarelations_search_contacts or mediarelations_search_sources instead.

mediarelations_create_saved_search

Creates a new standard search in Explore with a specified name and query. Supports boolean, keyword, and combined query types. Returns the ID of the newly created search.

listening_create_search

Creates a single new GenAI Lens prompt for your company (one prompt per call). topicId is required; a new prompt is always created filed under a topic: get an existing topic id from gail_get_filters (type: topics), or create one first with gail_create_topic if none fits; never invent or guess a topicId. (Existing prompts can later be unfiled via gail_update_prompt.) promptText is immutable after creation; to change an existing prompt's name, topic, labels, persona, or geo target use gail_update_prompt; to remove one use gail_delete_prompt (soft delete). The AI models the prompt runs against and its execution frequency come from your company's plan entitlements, not from parameters; they cannot be chosen here; a quota error means the plan's active-prompt limit is reached, not bad input. Never invent or guess IDs: take topicId from gail_get_filters (type: topics), labelIds from type: labels, countryCode from type: target_countries, personaId from type: personas. countryCode sets the prompt's target country (ISO 3166-1 alpha-2, e.g. 'US'). personaId, when set, tailors the response to that audience persona; a prompt has zero or one persona. Requires a user-scoped Authorization JWT; machine-to-machine tokens without a user claim are rejected. If a post-create step fails (for example label association or downstream event publish) the call still succeeds and transparency.notes explains. Returns one 'prompt_write_result' resource with status 'created', the new prompt's id and name, and a 'prompt' object echoing the persisted prompt exactly as it will appear in gail_get_filters (type: prompts); not date-scoped or paginated.

gail_create_prompt

Creates a single new GenAI Lens topic for your company (one topic per call). Call this before gail_create_prompt whenever no suitable topic already exists; gail_create_prompt requires a topicId, so check gail_get_filters (type: topics) first and only create a new one here if none fits. Only name is accepted: slug is generated server-side from name (not settable) and disambiguated automatically on collision. A name already used (exact, case-sensitive match) by an active topic in your company fails with a validation error: "A topic with this name already exists in your company. Choose a different name or reuse the existing topic from gail_get_filters (type: topics).", rather than returning or updating the existing topic; names differing only in case are allowed and receive a disambiguated slug. Requires a user-scoped Authorization JWT; machine-to-machine tokens without a user claim are rejected. Not for listing existing topics; use gail_get_filters (type: topics) for that; to rename a topic use gail_update_topic, and to delete an empty topic use gail_delete_topic; and it never files or moves a prompt into a topic; use gail_update_prompt's topicId for that. Returns one 'topic_write_result' resource with status 'created' and the new topic's id (pass this verbatim as topicId to gail_create_prompt or gail_update_prompt) and name (plus slug when available); not date-scoped or paginated.

gail_create_topic

Use this tool when the user wants to permanently delete an entire media list. This is a destructive operation that cannot be undone — the media list and all its contact/source associations will be removed for all users in the company. Returns a confirmation of the deletion. Always confirm with the user before deleting. Consider informing the user about the number of contacts/sources in the list before proceeding. Do NOT use this tool without first confirming with the user. Do NOT use this tool to remove individual contacts from a list — use mediarelations_update_media_list_contacts instead.

mediarelations_delete_media_list

Use this tool when the user wants to delete a saved search they no longer need. This is a destructive operation — the search will be soft-deleted, removed from the workspace, and permanently purged after 60 days. Returns a success message and the deleted search ID. This action affects all users in the company who may be using this saved search. Always confirm with the user before deleting. Do NOT use this tool without first confirming with the user. Do NOT use this tool to delete media lists — use mediarelations_delete_media_list instead.

mediarelations_delete_saved_search

Permanently deletes a standard (Explore) search by its ID. This cannot be undone.

listening_delete_search

Deletes a single GenAI Lens prompt by id. Destructive: confirm the user intends to delete this exact prompt before calling; there is no undo or restore via MCP. Never invent or guess IDs: take the exact 24-character hexadecimal ID from gail_get_filters (type: prompts). This is a soft delete: the prompt stops running, drops out of prompt listings, and frees one active-prompt quota slot for gail_create_prompt, but its historical AI model responses stay available to gail_get_responses and gail_get_analytics. Not for renaming, moving, or re-labeling a prompt; use gail_update_prompt for that. Requires a user-scoped Authorization JWT; machine-to-machine tokens without a user claim are rejected. Returns one 'prompt_write_result' resource: status 'deleted' with the prompt's id and name, or 'not_found' (a normal result, not an error) when no prompt with that id exists for your company, including one already deleted; the prompt is already gone, do not retry. Deleting a prompt owned by another company fails with an error instead. transparency.notes may report a non-fatal post-delete step (e.g. a delayed downstream event); the deletion itself still succeeded. Not date-scoped or paginated.

gail_delete_prompt

Permanently soft-deletes a single EMPTY GenAI Lens topic. Destructive and not undoable through this API, so confirm intent and the exact id (from gail_get_filters, type: topics) before calling; never invent or guess IDs. A topic that still contains prompts is refused with a validation error: 'This topic still contains N prompt(s). Topics must be empty before deletion: move each prompt to another topic with gail_update_prompt (topicId) or delete it with gail_delete_prompt, then retry.' This tool never deletes or moves prompts itself. Requires a user-scoped Authorization JWT; machine-to-machine tokens without a user claim are rejected. Returns one 'topic_write_result' resource: status 'deleted' with the topic's id and name, or status 'not_found' when the id does not resolve to a topic or it was already deleted (a normal result, not an error). Not for renaming (gail_update_topic) or listing (gail_get_filters, type: topics). Not date-scoped or paginated.

gail_delete_topic

Describe one source-aware filter, aggregation, or sort schema. Used by the Meltwater document retrieval and aggregation tools. Start at `/`, then follow returned child paths. When several paths are already known, request them concurrently in the same turn. The returned `schema` is a standalone Draft 2020-12 JSON Schema fragment; `children` contains immediate navigation links. Avoid `depth` greater than 1 at a root because recursive schemas can exceed the response budget; increase depth only on a targeted child path. A child's `description` under `children` is a compact catalog summary for choosing a path, not its full contract. Follow the child's own `path` and read the loaded `schema.description` before constructing a filter or aggregation from it; the two commonly differ. This tool returns schema documentation only; it does not retrieve or aggregate Meltwater documents.

content_describe_schema

Describe one owned-content aggregation, filter, or sort schema. Used by `owned_posts_retrieve`, `owned_posts_aggregate`, and `owned_pages_aggregate`. `pages` supports `filters` and `aggregations`; `sort` applies only to `posts`. Start at `/`, then follow returned child paths. The returned `schema` is a standalone Draft 2020-12 JSON Schema fragment; `children` contains immediate navigation links. A child's `description` under `children` is a compact catalog summary for choosing a path, not its full contract; follow the child's own `path` and read the loaded `schema.description` before constructing a filter or sort from it. This tool returns schema documentation only; it does not retrieve owned content.

owned_content_describe_schema

Reads raw stored AI model responses: full text, quotes, and citations (the evidence layer). Provide exactly one of responseId or query. responseId fetches one full stored response (its complete text, plus model and prompt IDs); it does not return citation links or a prompt object, so it cannot recover fuller citations than a keyword row's top links. Use the responseId value from a prior gail_get_responses result, or its bare id portion (a 24-character hexadecimal ObjectId or a UUID); never invent or guess IDs; an unknown ID fails with 'Response not found', not an empty result. query with matchType 'keyword' (default) is exact text search over stored response text (not prompt text), paginated, returning compact rows (a truncated response snippet and top citation links, not the full text; fetch responseId only for the untruncated response text). query with matchType 'semantic' is meaning-based search over response sentences; use it when exact wording is unknown or keyword is too narrow, capped at limit groups instead of paginated. minScore applies only to semantic; cursor, sortBy, and sortOrder only to keyword; limit applies to both, defaulting to 5 for keyword search; wrong combinations are rejected. Omit dateRange to default to the last 7 UTC calendar days; the applied window is echoed in transparency.dateRange, so never present defaulted results as all-time. Never invent or guess IDs or names: take promptIds/promptNames from gail_get_filters (type: prompts), modelNames from type: models. Returns 'llm_response' resources (responseId/keyword) or 'response_sentence' snippet groups (semantic); an empty resources array means nothing matched in the window, not an error. Not for aggregated metrics and rankings (use gail_get_analytics) or pre-generated weekly advice (use gail_get_recommendations).

gail_get_responses

Fetches a brand's pre-generated AI visibility recommendations, produced by a weekly analysis run; not computed on demand. Every stored run is scoped to one topic, and each call searches a 15-day lookback window ending at endDate (default: today, UTC) for the most recent run per topic whose analyzed week starts in that window (the exact window searched is named in transparency.notes, not transparency.dateRange). Narrow with topicIds and/or topicNames to get the latest run for those specific topics; omitting both returns the latest run for the 5 largest topics by active prompt count instead of all topics, named in transparency.notes; to cover different or more topics, list their topic IDs explicitly (from gail_get_filters type: topics). Provide exactly one of brandId or brandName. Never invent or guess IDs: take brand IDs and exact names from gail_get_filters (type: brands), topic IDs from type: topics. Each row carries category: CITATIONS (citation/source authority: which domains get cited, including the brand's own), NARRATIVE (how models frame and describe the brand), SHARE (share of voice/visibility versus competitors), or PROMPTS (prompt coverage gaps); plus title (plain text), diagnosis, recommendation, evidence, topic, topicId, runDate, actions, dismissed, and optional owner, dueAt, effort, impact, and status. Actions contain a title and checked flag; diagnosis/recommendation/evidence may carry markdown-linked URLs, title never does. Rows are deduplicated on (category, evidence), keeping the first occurrence. transparency.dateRange echoes the analysed week(s) the returned rows actually cover (a union across topics when they last ran in different weeks; see transparency.notes), never the 15-day lookback window itself. An empty resources array means no run landed in the searched window for any selected topic, or every candidate recommendation fell below the server-side confidence threshold; not an error; transparency.notes names the window searched, and retrying with an earlier endDate reaches an older run rather than repeating the same call. No pagination: all rows (at most 50) arrive in one call. Not for raw stored response text and quotes (use gail_get_responses) or aggregated metrics and rankings (use gail_get_analytics).

gail_get_recommendations

Runs one aggregated GenAI Lens analytics report per call from stored AI responses. Fifteen reports: brand_metrics, model_comparison, top_prompts, brand_analysis, competitive_trends, prompt_visibility_gaps, sentiment, attributes, trends, topics, citations, citation_peaks, brand_source_share, journalists, influencers. Not for raw text (gail_get_responses) or advice (gail_get_recommendations). Resolve names/IDs via gail_get_filters first; never invent them. Parameters are report-specific and validated at call time; an invalid or missing one names the valid set; for a given report's required/optional parameters, defaults, and metric scale, call gail_get_filters (type: reports) rather than guessing from this description. Omitting dateRange defaults to the last 7 UTC days (see transparency.dateRange; never all-time). model_comparison and top_prompts use different visibility scales (0-100 vs 0-100.00); never compare directly. One resources[] entry, kind analytics.<report>.

gail_get_analytics

Discovery tool for every other gail_* tool: call it first to obtain exact IDs and names; never invent or guess IDs. Provide exactly one of type or types. type selects one inventory per call: models = AI models your company's plan is entitled to (name or slug usable as modelNames); brands = your company's own tracked brand profiles; these exact names are the only valid values for brand/brands/competitors/brandName (id → brandId); brands configured only as nested competitors on a profile are not listed and cannot be named in any parameter (gail_get_analytics includeCompetitors=true, on report brand_analysis, competitive_trends, prompt_visibility_gaps, or attributes, includes them automatically); each row's primaryColor and secondaryColor, when set, are the brand's colors (uppercase #RRGGBB) for charts; prompts = your active prompts (id → promptIds; promptText is truncated to 300 characters; each row's labelIds and personaId, when present, reuse verbatim as labelIds and personaId elsewhere); topics = groupings used to organize prompts (id → topicId/topicIds); labels = non-hidden labels (id → labelIds); target_countries = prompt geo targets whose id is the ISO 3166-1 alpha-2 code (e.g. 'US'); personas = active personas (id → personaIds; each row's color, when set, is the chart series/settings-list color); reports = a static fourteen-row catalogue of every gail_get_analytics report (id/name = report name; question, requiredParameters, optionalParameters, defaults, and metricScale describe how to call it), not scoped to your company. types accepts several of the above in one non-paginated call (limit applies per type; cursor is rejected alongside types; a type whose results don't fit is named in transparency.notes with a call-again hint); prefer type with cursor when you need to page through one inventory fully. Prompt, topic, label, persona, and brand IDs are 24-character hexadecimal IDs (e.g. '665f1c2ab3d4e5f6a7b8c9d0'). searchText narrows by name (brands also match aliases; prompts full-text over name and prompt text; reports match name or question); omit it to list everything. semanticQuery is a meaning-based alternative to searchText for prompts only (type=prompts or types=['prompts']): use it when the right prompt is unknown by exact wording (e.g. 'prompts about pricing complaints') rather than a keyword match. Mutually exclusive with searchText and cursor; results are capped at limit and ordered by relevanceScore descending, not paginated. additionalFilters still narrows a semanticQuery call the same way it narrows searchText. additionalFilters adds structural, server-side narrowing per type (prompts: topicIds, labelIds, countryCodes, personaIds, promptIds, status; topics: slug, includeDeleted, created/updated date bounds; models: slugs, includeInactive; labels: includeHidden; personas: includeArchived) and is how you answer questions searchText cannot, such as prompts with no label (labelIds: ['NONE']), no geo target (countryCodes: ['NONE']), or no persona (personaIds: ['NONE']); brands, target_countries, and reports accept no additionalFilters, and a block for a type you did not request is rejected. Listings are current inventory with no date filter; an empty resources array means nothing matched, not an error. limit defaults to 25 (max 100, or 200 for a single type=prompts call). To page a single type, pass transparency.nextCursor back verbatim as cursor with the same type and searchText; a missing nextCursor (or hasMore=false) means the end. Not for raw stored response text and quotes (gail_get_responses), aggregated metrics and rankings (gail_get_analytics), or pre-generated weekly advice (gail_get_recommendations).

gail_get_filters

Use this tool when the user wants to list, browse, or search their media lists. This is a read-only operation scoped to the authenticated user's company and workspace. Returns a paginated response with: count (total number of media lists) and hits (array of media list objects with name, key, journalistCount, creation date, and workspace membership). Supports pagination (page/size), name search (q parameter), sorting (sort: NAME_ASC, NAME_DESC, JOURNALISTCOUNT_ASC, JOURNALISTCOUNT_DESC, CREATED_ASC, CREATED_DESC), and label filtering (labelId). Do NOT use this tool to search for contacts within a media list — use mediarelations_search_media_list_contacts instead. Do NOT use this tool to modify media list membership — use mediarelations_update_media_list_contacts or mediarelations_update_media_list_entries instead.

mediarelations_get_media_lists

Use this tool when the user wants to inspect a specific outreach — its full email body, subject, or metadata. Typical user requests: "show me the content of this outreach", "what does this campaign email say?", "get the details of distribution X". This is the preferred tool for general outreach inspection. This is a read-only operation scoped to the authenticated user's company. Returns the complete email body (HTML), subject line, sender address, and associated metadata. Do NOT use this tool for campaign performance metrics — use mediarelations_get_outreach_counts instead. Do NOT use this tool for overall open rate and delivery counters — use mediarelations_get_outreach_counts instead. Do NOT use this tool for per-recipient engagement data — use mediarelations_get_outreach_recipients_aggregations instead.

mediarelations_get_outreach_content

Use this tool when the user wants delivery and engagement analytics counts for a specific outreach distribution. This is a read-only operation scoped to the authenticated user's company. Returns a single object with counters for delivered, sent, open, click, bounce, blocked, unsubscribed, untracked, and replied events, plus derived rates (openRate, ctoRate, unsubscribeRate, replyRate). The response shape varies slightly depending on the analyticsType: 'all' returns the full object, 'open' focuses on open-related metrics, and 'click' focuses on click-related metrics. Do NOT use this tool for per-recipient breakdowns — use mediarelations_get_outreach_recipients_aggregations instead.

mediarelations_get_outreach_counts

Use this tool when the user wants to see engagement details or analytics per recipient for a specific outreach distribution. Typical user requests: "show me the analytics per recipient", "who opened my outreach", "show me the recipient breakdown for this outreach". This is a read-only operation scoped to the authenticated user's company. Returns a paginated result using the page (default 0) and pageSize (default 100) parameters. The response contains: a contacts array (individual recipients with emailAddress, status — one of: opened, unopened, bounced, replied, untracked — bounce type, unsubscribe flag, and profile data), a sources array (journalist/source contacts with the same fields plus name and location), and a count with the total number of recipients. Do NOT use this tool for campaign-level metrics — use mediarelations_get_outreach_counts instead. Do NOT use this tool for delivery counters and rates — use mediarelations_get_outreach_counts instead.

mediarelations_get_outreach_recipients_aggregations

Use this tool when the user wants to see a list of their outreach campaigns. Typical user requests: "show me all my outreaches", "list my campaigns", "what outreaches do I have?". Returns a paginated result — use page and pageSize to navigate results, and sortField / sortOrder to control ordering. This is a read-only operation scoped to the authenticated user's company. Returns an object with a total count and a hits array of outreach items, each containing subject, status, sender address, creation date, and associated media lists. Do NOT use this tool to get the full details of a specific outreach (like full body) — use mediarelations_get_outreach_content instead.

mediarelations_get_outreaches

Use this tool when the user wants to inspect the full configuration of a specific saved search by its ID. This is a read-only operation scoped to the authenticated user's company. Returns the full saved search object including searchConfiguration, relevantConfiguration, searchType, keywordType, and metadata (creator, dates). Use the optional 'fields' parameter to limit the response for better performance. Use the returned searchConfiguration to execute the saved search with mediarelations_search_contacts or mediarelations_search_sources. Do NOT use this tool to list all saved searches — use mediarelations_get_saved_searches instead.

mediarelations_get_saved_search

Use this tool when the user wants to list or browse their saved searches. This is a read-only operation scoped to the authenticated user's current workspace. Returns a paginated response with: hits (array of saved search metadata including name, type, creator, dates), count, page, size, and totalPages. Supports pagination (size default 12, page 0-indexed), partial name filtering (search parameter, case-insensitive), sorting (sortBy: searchName/createdAt/updatedAt/executedAt, sortOrder: asc/desc), and field selection (fields parameter, comma-separated). Do NOT use this tool to retrieve full details of a specific saved search — use mediarelations_get_saved_search with the saved search ID instead.

mediarelations_get_saved_searches

Fetches slices of one shared skill. id comes from meltwater_list_skills, meltwater_search_skills, or meltwater_create_skill. Platform skills outside the company's entitled tool packages are not found. Tenant skills the caller can read are unchanged. Default include=['header'] returns metadata only. Add 'content' for the SKILL.md file map — large payload, use only after the user picks a skill; that payload is terminal (follow SKILL.md, do not chain it). Add 'changelog' for version history written when meltwater_update_skill replaces files. header and content are null when the skill is missing or not accessible.

meltwater_get_skill

Returns the details of a single standard (or Explore) search by its ID, including its name, query type, query configuration, and metadata.

listening_get_search

Use this tool when the user wants to see all available labels for their company. Typical user requests: "show me all labels", "what labels do I have?", "list all tags I can use for my media lists". This is a read-only operation scoped to the authenticated user's company. Returns an array of label objects, each with an id and name. Do NOT use this tool to assign or remove labels from a media list — use mediarelations_update_medialist_labels instead.

mediarelations_get_labels_by_company

Use this tool when the user mentions a specific influencer by name or handle and you need to find their profile. Returns a list of matching profiles sorted by relevance, with influencer ID, name, and social media handle. Pass the returned handle, exactly as given, with the same network to get_profile_overview, get_profile_demographics, or get_profile_content for detailed data.

klear_get_influencer_handles

Use this tool when the user wants to see an influencer's posts or content. Returns a list of posts with textual content, publish date, engagement metrics, and the social network. Do NOT use this tool for profile-level stats — use get_profile_overview instead.

klear_get_profile_content

Use this tool when the user wants to understand an influencer's audience composition. Takes an Instagram handle and returns that profile's Instagram audience demographics — age distribution, gender split, top countries, cities and states, skills, brands and hashtags. Only Instagram is supported; there is no equivalent for other networks. Any of the demographic breakdowns may be missing when there is not enough audience data. Do NOT use this tool for the influencer's own profile info — use get_profile_overview instead.

klear_get_profile_demographics

Use this tool when the user wants a summary of an influencer's profile. Returns influence scores, network statistics (total audience, per-network follower/subscriber counts), engagement metrics and rates, audience highlights (top country, skills, reach), follower/engagement/reach history, recent post counts, predicted EMV, predicted pricing, the creator's display name and bio, and an AI-generated overview of the creator and their content. Do NOT use this tool for audience demographics — use get_profile_demographics instead.

klear_get_profile_overview

Use this tool when the user provides a free-text description of the type of influencer they are looking for (e.g., "vegan food bloggers in NYC with over 50K followers"). This is a Klear influencer-marketing tool, scoped to paid influencer/brand-partnership data — audience, pricing, and EMV — not organic/earned-media listening authors. Returns a list of influencer profiles matching the query, each with name, per-network usernames, follower counts, and influence/engagement scores. Supports filtering by multiple countries (e.g., countries: ["US", "UK"]). Supports filtering by multiple networks (e.g., networks: ["instagram", "youtube"]); defaults to instagram when omitted.

klear_influencer_search_by_prompt

Use this tool when the user wants to see which brands an influencer has worked with before. Returns a list of past brand collaborations, each with the brand name and the influencer's posts mentioning that brand.

klear_list_profile_collaborations

Returns custom fields defined in the user's Explore+ account. Custom fields are user-defined categorisations (tag fields or multivalued fields) applied to documents in the Explore+ index. Requires a workspace_id — use account_get_workspaces to look up IDs, or pass none for account-level fields. Results are paginated (default 10 per page, max 100): use page and page_size, and compare against total to fetch the remaining pages.

explore_plus_list_custom_fields

Returns all saved filter sets in the user's account. Filter sets are reusable combinations of filters (source, language, country, etc.) applicable to searches and analytics. Each result includes the filter set ID and configuration.

listening_list_filter_sets

Returns optimised searches in the user's Explore+ account, which are distinct from standard (Explore) searches returned by listening_list_searches. Optimised searches define what content is collected into the Explore+ private index. Requires workspace_id; use 'none' for account-level access. Supports optional folder filtering, pagination, and show_hidden (set to true to include searches within hidden folders).

explore_plus_list_searches

Lists shared skills the current user can access on the MCP surface (MCP-only plus unrestricted/empty surfaces). Platform skills (share level all) are only those in a tool package the company is already entitled to. User, company, and workspace skills follow their share rules and are not package-gated. Use this to browse or page through skills. Do not use it for a natural-language lookup — use meltwater_search_skills, which ranks by meaning. Default page=1, limit=10, sortOrder=desc; a smaller limit hides later pages. Returns headers only (id, name, description, trigger, share_level, create_ts). Each results[].id is the id for meltwater_get_skill, meltwater_update_skill, and meltwater_delete_skill. Fetch files with meltwater_get_skill include=["content"] only after the user picks a skill (content can be large).

meltwater_list_skills

Returns all standard (Explore) searches in the user's account. Does not include Explore+ optimised searches (the custom-index collection definitions) — those come from explore_plus_list_searches. Each result includes the search ID, name, and query. Optional filters: query_type (boolean, keyword, combined) and expand_combined.

listening_list_searches

Use this tool when the user wants to save an outreach as a draft without sending it. Typical user requests: "save this as a draft", "create a draft outreach", "I want to prepare this email but not send it yet". All fields are optional — an empty draft can be created with no parameters. To associate media lists, pass mediaListIds — the recipients are resolved automatically from those lists (you do NOT pass recipients). Returns a confirmation message on success. Do NOT use this tool to send or schedule an outreach — use mediarelations_schedule_outreach instead.

mediarelations_create_outreach_draft

Use this tool when the user wants to permanently delete an outreach. Typical user requests: "delete this outreach", "remove this campaign", "I want to delete distribution X". Requires the distribution ID. This action is irreversible. Returns a confirmation message on success. Do NOT use this tool to modify a scheduled outreach — use mediarelations_update_outreach instead.

mediarelations_delete_outreach_distribution

Use this tool when the user wants to generate a personalized AI pitch for a specific journalist or contact. Typical user requests: "generate a pitch for this journalist", "write a personalized message for John Smith", "create a tailored pitch for this contact". Requires the journalist's ID; optionally accepts a draft email (userDraftContent) or a key message to guide the pitch. Returns the generated pitch text that can be used or refined in an outreach email. Do NOT use this tool to send the pitch — use mediarelations_schedule_outreach or mediarelations_create_outreach_draft instead.

mediarelations_generate_outreach_pitch

Retrieve individual Meltwater documents matching an optional Meltwater Boolean query and structured filters. Use this tool when Meltwater documents must contain explicit terms or satisfy a Meltwater Boolean query. It is also the retrieval tool for a Meltwater customer's custom index (the Explore+ index) when `source="custom"`. The result contains individual Meltwater documents. `sort` biases the results, so they are not necessarily ordered as requested. Do not use this tool to count or break down the complete matching set; use `content_aggregate` instead.

content_retrieve_keyword