Tool reference

Answer "which of my projects needs an article right now?" across every project this connection can access. Each project is analysed against its own publishing rhythm (or its automation routine) and ranked by urgency: how long since its last article, how late that makes it, what is still in its pipeline, and what to do next. Returns a compact ranked list plus a summary of how many projects are overdue, due, covered or on track. Call get_publishing_cadence with one project_id for the full detail of a single project.

Report exactly what this connection can and cannot do: the app it is acting as, its granted permissions, the projects it reaches and your role on each, which tools are usable (and why the others are not), and where the account owner can review every call made through it. Call this when a tool was refused, before attempting anything destructive, or when the user asks what you are allowed to do here.

No arguments.

List the projects this connection can access. Use the returned project_id to target other tools when more than one project is available. project_url is the live published site; dashboard_url is the editor.

No arguments.

Fetch one background job (worker-request) by id, with its status, recent logs and final result or error. Use the job_id returned by list_jobs, translate_article or translate_missing.

List the AI agents configured for a project (translation, article-idea, recommendations, article-writer, image, cover, tweet-to-article, emphasis) with their status. For the translation agent, languages lists the target languages it can translate into. For an image agent, model is what draws article covers - pass its agent_id to generate_image, or to generate_article to choose which one illustrates the piece. For a tweet-to-article agent, thread_rules is how it reads an X thread and what it asks the article to be - pass its agent_id to tweet_to_article, scrape_tweet or analyze_tweet_discussion to read with those rules. For a cover agent, cover_rules is what the picture at the top of an article looks like and where it comes from - pass its agent_id to create_cover. A project can hold more than one of a type, so several cover agents can come back: the one a thread article is illustrated by is named in the tweet-to-article agent thread_rules as cover_agent_id, and configure_agent with create_new adds another. For an emphasis agent, emphasis_rules is how much italic the articles carry; a project with no such agent still gets the built-in light touch, so the absence of one here is a default and not an off switch. The same is true of the cover agent: articles are illustrated with the built-in defaults when there is none. Use this before launching work with translate_article, translate_missing or generate_image. To create or change a writer, image, cover or tweet-to-article agent, use configure_agent.

List the image models this account can generate with, to pass as `model` to generate_image. Includes every OpenRouter slug the model library carries (Gemini, GPT-image, FLUX, Seedream, Grok Imagine and the rest of the gateway) alongside the direct-provider models. Badges tell them apart: recommended is the product default, fast is cheap and quick, premium is the expensive high-detail tier. Filter with provider or search when the list is long.

List background jobs (worker-requests) for a project, newest first, with their status (pending, processing, completed, failed, cancelled). Filter by type or status to follow work launched with translate_article, translate_missing or generate_recommendations. Use get_job for the full detail of one job.

Get a single-call overview of a project: its languages, article counts (total, by status, by language), idea backlog, and - when allowed - its active AI agents, its in-flight job queue and a setup block saying whether the writer and image agents are configured and whether X threads have house rules. Call this first to orient before using more specific tools.

No arguments.

List the published articles most worth updating rather than replacing: the oldest untouched ones, scored up by thin content, missing meta title or description, missing cover and missing internal links. Each candidate comes with a refresh score out of 100 and the reasons behind it. Word counts are approximated from the stored body length.

Fetch one article by id or by handle. Returns the full document including markdown body. The article must belong to the targeted project.

List the change history for one article, newest first. Each entry is a recorded mutation (create, update, edit, delete) with its source and a size delta. Use the returned history_id with rollback_article to restore a previous version. Set include_content to also return the before/after markdown of each version (heavier).

Show the editorial calendar of a project around today: articles published in the recent window, everything booked ahead (scheduled articles and drafts with a future date), upcoming automation routine runs, and the days in the horizon with nothing planned. Flags booked slots whose article is still missing a cover or a meta description, because those cannot publish as-is. Dates are given both as instants and as local days in the project timezone.

Find what a project is NOT covering: languages that lag behind the main one, categories declared in the project but never used (or untouched for months), published articles with no category at all, categories used on articles but missing from the project settings, and the share of the catalogue that has not been touched in a long time. Returns per-language and per-category coverage plus a prioritised list of gaps, each naming the tool that closes it. Pair with get_publishing_cadence: that one answers when to publish, this one answers about what.

Answer "when should I publish the next article here?" for one project. Returns the real publishing rhythm (median gap between articles, articles per week/month, trend, usual weekday and hour), how long ago the last article went out, what is still in the pipeline (drafts, scheduled, review queue, ideas), and a verdict: on_track / due_soon / due_now / overdue / covered / no_history, with a due date, an urgency score out of 100 and the concrete next actions to take. Use find_projects_needing_articles instead when the question spans several projects.

Check whether cache purges are reaching the customer's site: whether revalidation is configured for the project, the recent delivery attempts with the site's HTTP answer, how many are waiting to be retried, and - when article_id is given - whether that article still owes a purge (its body changed more recently than the last purge we delivered, so the next automation tick will send one). Start here when someone reports that the live site shows an old title, an old body or a deleted article. Read-only: it never calls the site, revalidate_article does that.

List AI-generated article ideas for the targeted project. Optionally filter by status (new, approved, dismissed, promoted). Returns ideas plus per-status counts.

List every picture in one article: its cover, plus each image in the body with a position number, its alt text, caption, and which section it sits in. The position numbers are what add_article_image, replace_article_image, remove_article_image and move_article_image take, so call this first when the user says "the second photo" or "the image in the pricing section". Also flags images that are still hosted outside the project, which break when the other host does.

List articles in a project. Supports pagination, search by title/handle, optional language and category filters, and date range.

List the categories declared for a project, with how many articles each holds. These are the values create_article and update_article expect in `categories` and list_articles in `category`: pass one of these rather than inventing a label, because an undeclared category still saves but the dashboard filters never offer it. Also returns the categories found on articles that the project never declared, which is what get_content_gaps flags as a taxonomy gap.

No arguments.

Read an X (Twitter) post, the posts it is built on, and the discussion under each of them, ranked. Returns the post, the author's own follow-ups, and the replies that are actually worth reading - scored on engagement, substance, whether they push back with numbers or first-hand experience, and whether the author stopped to answer them. Promoted posts injected into the reply timeline are removed, so what comes back is the real conversation. By default the read goes one hop back: when the post quotes another post, answers one, or links to one, that post is read too and comes back in `sources` with its own replies, because a post commenting on another one usually leaves the numbers and the original claim in the post underneath. Also returns `dossier`: the whole thing compiled into one readable brief, which is exactly what analyze_tweet_discussion and tweet_to_article feed to the model. Also reports the picture the post carries (`cover_candidate`), every picture in the thread that would go inside the article (`images`), the video inside it (`video`), and the pages the posts link to (`links`), so what the article will be built from can be shown to the user before anything is spent. Free (no AI credits, no article created). Use it to decide whether a thread is worth writing about, or to quote what people said.

Propose the concrete dates for the next articles of a project, ready to pass to schedule_article. Slots follow the project's real rhythm (or its automation routine), reuse the weekday and hour it usually publishes at, are expressed in the project timezone, and skip any day that already has an article published or booked. Returns the slots plus the cadence verdict they were derived from.

List the files in a project's asset library: everything uploaded from the dashboard plus every image generated for this project (article covers, in-article images, anything drawn over MCP). Each one comes back with a durable public URL you can embed in markdown or set as an article cover - check here before generating a new image, the picture may already exist. Filter with search (matches the file name and the prompt an image was generated from), source, or article_id.

Read an X thread, and the posts it is built on, and say what is actually going on. Returns the claim the post made, the disagreements and how each was answered, the quotes worth lifting with their handles, the observations that only become visible once the whole thread is read, the questions nobody settled, and a suggested headline, angle, keywords and cover prompt. When the post quotes or answers another one, that post and its own reply section are read too, and the analysis adds `source_chain`: what the original actually said, and whether the post on top represents it fairly. That gap is usually the article. Use it to judge whether a thread is worth publishing and to agree an angle with the user BEFORE writing. Spends AI credits (one model call). Writes nothing: call tweet_to_article when you want the article.

Cancel a background job that is still queued (status pending), for example a translation launched by mistake. Jobs already processing, completed or failed cannot be cancelled. Get the job_id from list_jobs.

Set up or update a project AI agent - the article writer (its audience, tone and house rules), the image agent (its art direction), the cover agent (what the picture at the top of every article looks like and where it comes from), the tweet-to-article agent (how an X thread is read, whether its pictures and the pages it links to go into the article, and what the piece should be), or the emphasis agent (how much italic the articles carry). Call it with only agent_type to GET the questions to ask the user, together with what the project already says about itself so you can propose answers instead of interrogating. Call it again with the answers as fields to save them. A preset can be applied instead with template. A project can hold SEVERAL agents of a type: create_new adds another one and returns its agent_id, which is how a blog gets a second cover style for the articles it makes from X threads (point the tweet-to-article agent at it with cover_agent_id) without changing the covers of everything else. This is what generate_article points you to when a project has no configured writer yet, where create_cover gets the look of a cover from, and where tweet_to_article gets its reading depth and editorial rules from.

Write a brand-new article from a topic you provide. Creates the draft immediately and queues the AI writer to fill it in, so the article appears in the dashboard right away with a "writing" state. The same job also draws the cover and links it to the article, so no second call is needed and no article comes out unpublishable for want of one; the project image agent supplies the style when it has one. Set generate_cover false to skip it, or image_agent_id to pick which image agent draws it. Bring your own photos with images: a list of public URLs, which are copied into the project CDN, placed in the body by the writer where they fit the text, and used as the cover (the first one, unless you flag another) instead of drawing one. Returns article_id + job_id - poll get_job until it succeeds, then read the result with get_article. Spends AI credits. IMPORTANT: when the project has no configured writer yet, this writes NOTHING and returns status "setup_required" with questions to put to the user - ask them, save the answers with configure_agent, then call this again (or pass skip_setup true to write with a generic default). To review the topic first instead, use create_idea + promote_idea.

Generate an AI image for one project and get back a durable URL you can use as an article cover or embed in markdown. The image is filed in that project's asset library, so it also shows up on the Assets page of the dashboard and can be reused or removed later (list_assets / delete_asset). An image therefore belongs to a project: when this connection reaches more than one, ASK the user which project it is for and pass project_id - never pick one yourself, because it spends that project's credits and lands in its library. Runs through the project image agent by default: its model and art direction are applied, so the result matches the rest of the blog. Pass agent_id to pick a specific image agent, or override provider/model/size/quality yourself - with provider "openrouter" the model is any OpenRouter image slug ("google/gemini-3-pro-image", "black-forest-labs/flux.2-pro"); call list_image_models to see what is available. Pass article_id + set_as_cover to attach the result to an article in one call (also needs the "Edit articles" permission). Spends AI credits.

Queue a recommendations run (related-articles suggestions) for every published article in the project. Articles already queued are skipped. Spends AI credits per article when the worker runs. Requires an active recommendations agent.

Promote an article idea into a draft article and queue its AI generation (the same "Accept & write" action as the dashboard). Returns the created article id and the generation job id (track it with get_job). Spends AI credits when the writer runs. Idempotent: promoting an already-promoted idea returns the existing job. When the project has no configured writer yet, this promotes nothing and returns status "setup_required" with questions to ask the user - see configure_agent, or pass skip_setup true.

Queue a translation of one article into one or more target languages. Returns immediately with queued job ids (track them with list_jobs / get_job). Each translation lands as a DRAFT unless publish is true, in which case it goes live by itself as soon as it is done. Target languages must be languages the project is configured for (see describe_project); languages already translated or already queued are skipped. Each queued language spends AI credits when the worker runs it. Requires an active translation agent in the project (see list_agents).

Queue translations for every missing (article, language) pair in the project. Skips pairs already translated or already queued. Each translation lands as a DRAFT unless publish is true, in which case every translation of an already-published article goes live by itself as soon as it is done. Set dry_run to see how many translations would be created (and which languages) without queuing anything. Each queued translation spends AI credits when it runs. Requires an active translation agent.

Turn an X (Twitter) thread into an article. Reads the post, the post it is built on when it quotes or answers one, and the discussion under each, ranks the replies that matter (the ones with numbers, first-hand experience, real pushback, and the ones the author answered), works out what the whole chain adds up to, and hands all of it to the project writer as source material with the instruction to quote real people by handle and to say what the original post claimed before what the post on top made of it. Creates the draft immediately and queues the writer, so it appears in the dashboard right away in a "writing" state, and it always comes out with a cover: the same job draws one from a prompt written for this thread, so nothing else is needed before publishing. The pictures the posts carry - the charts, the benchmark tables, the screenshots, including the ones posted in the replies - go INSIDE the article by default: each one is copied into the project CDN, described by a vision model and placed by the writer where it belongs, because they are the evidence the discussion is about (pass thread_images false for a text-only piece). The pages the posts link to are opened and read the same way, so the article is written from the paper, the announcement or the benchmark itself rather than from the sentence somebody wrote above the link (read_links false skips them). When the post carries a picture worth showing - a chart, a screenshot, a demo video - pass cover_source "tweet" and the cover becomes that photo, or the frame from the video, instead of a drawing. When the post IS a video and the user wants the article to show it, pass video_frames: stills are cut from inside the video, read by a vision model and placed in the body by the writer, and what happens in the video goes to it as source material. Returns article_id + job_id - poll get_job, then get_article. Spends AI credits. Use analyze_tweet_discussion first when the user should approve the angle: this call reuses that read and that analysis instead of paying for them twice. Refuses a thread this project has already turned into an article in the same language, and names the existing one, so the same link pasted twice never quietly becomes two pieces (pass allow_duplicate to write a second take anyway). IMPORTANT: when the project has no configured writer this writes NOTHING and returns status "setup_required".

Create an article from content you wrote yourself (no AI credits spent). The handle, slug and URL section are derived from the title and the project's URL structure unless you pass an explicit handle. Publishing straight away requires a cover and a meta description, exactly like publish_article - otherwise it is created as a draft.

Add an article idea to the project backlog. Use it to bring your own topic into Reblog: the idea shows up in the Ideas board for review, and promote_idea turns it into a written article. Free (no AI credits). Duplicate titles are not inserted twice - the existing idea is returned instead.

Insert one photo into the body of an article from a public URL (or an asset_id from list_assets), at the position you choose: at the end, under a section heading, after a numbered paragraph, after a sentence you quote, or next to an existing image. The file is copied into the project CDN first, so the article never points at someone else's host - pass rehost false to link the original URL as-is. Give it alt text (good for SEO and accessibility) and an optional caption. Use dry_run to see where it would land before saving. To place several photos at once, use attach_article_images; for the cover, set_article_cover.

Make targeted edits to an article body by exact find-and-replace, without resending the whole markdown. Read it first with get_article, then pass one or more { old_string, new_string } edits. Each old_string must appear exactly once unless replace_all is set. Set dry_run to validate and preview the result without saving. Each applied edit returns a context snippet plus a preview_url to view the result.

Decide on an article awaiting approval in the review queue. decision "approve" publishes it immediately (internal + configured CMS, fails without cover/meta description); "reject" returns it to draft and cancels any auto-publish deadline. Only works on articles with status awaiting_approval.

Put several photos into an article at once, from public URLs (or asset_ids). They are copied into the project CDN and spread through the body - one under the opening paragraph of each section that has no picture yet, in the order given - and the first one becomes the cover when the article does not have one. Use it when someone hands over a batch of photos for an article. placement "end" instead puts them all at the bottom in order; set_cover_from controls the cover. Use dry_run to see the plan first, and add_article_image when one photo must land in one exact spot.

Draw the cover image of an article and attach it, in one call. The picture is about the article, not about its title: a short pass reads the piece and picks what should be in the frame, then the project's cover agent decides what it looks like (list_agents shows it as cover_rules), so every cover on a blog comes out of the same art direction. This is what to call when publish_article says an article has no cover, or when someone asks to add or replace one. Pass subject to say what the picture should show, art_direction for a one-off look, or prompt to send an exact prompt of your own and skip both. When the article was made from an X thread and the rules say the cover comes from the source material, the picture in the post is used instead of a drawn one - the chart, the screenshot, the frame from its video - which costs nothing at all; source overrides that for one article. A picture taken that way goes out on a card carrying the article title and crediting the handle it came from, because a screenshot on its own says nothing about the piece (the cover agent's title_overlay turns that off). Use preview to see the rules, the model and what it would do for free before spending anything. An article that already has a cover is left alone unless replace is true. Spends AI credits when it draws.

Add a light touch of italic emphasis to an article that reads flat, or strip the italics it already has. The typographer picks a few terms - the concept a sentence introduces, the two or three words it turns on - and italicises those and nothing else: it may only add asterisks, never change a word, so a bad suggestion is dropped rather than written. How much is the project's own house style, held by its emphasis agent (list_agents shows it as emphasis_rules), so pass intensity only when this one article should differ. Articles written from now on already come out formatted this way; this is for the ones written before. Use preview to see what would change without touching the article or spending anything, and mode "clear" to remove every italic instead.

Move a photo that is already in an article to another place in the body, keeping its alt text and caption. Target it by index (from list_article_images) or URL, and give the new position the same way add_article_image does: end, top, after_heading, after_paragraph, after_text, or before_image / after_image. Note that before_image / after_image count the images as they are once this one is lifted out. Use dry_run to preview.

Publish an article (set its status to published). Fails if the article has no cover image or no meta description, exactly like the dashboard. Also refreshes the linked translations and the recommendation references that point to it. Idempotent: publishing an already-published article is a no-op.

Remove one photo from the body of an article, together with its caption line. Target it by index (from list_article_images) or by URL. The file stays in the project asset library so other articles and translations keep working - use delete_asset if it should be gone for good. Use dry_run to check which image would go.

Swap one photo in an article for another, or fix its alt text or caption, without touching the rest of the body. Target it by index (from list_article_images) or by its current URL. Pass url (or asset_id) to change the picture, alt/caption to change the wording - any combination. A new URL is copied into the project CDN first unless rehost is false. Use dry_run to preview.

Force the customer's site to drop its cached copy of an article, now. Publishing, editing, renaming and deleting already do this on their own, so reach for it when a page on the live site is still showing something we changed - typically because the site was unreachable when the purge first fired. Call it with no article_id to send a test ping instead, which checks the endpoint and the secret without touching an article. Answers delivered true/false: false with a `reason` means revalidation is not configured for this project (nothing was sent, nothing is queued), false with an `error` means the site was reached and refused. Use get_revalidation_status to see the delivery history.

Restore an article body to a previous version recorded in its history. Get the history_id from get_article_history. By default this restores the "before" snapshot (undoing that change); set side to "after" to re-apply it. The rollback is itself recorded in history, so it can be undone too.

Book an article to publish itself at a future date, or clear an existing booking. Sets the article to "scheduled" with a publish_at instant; when the slot comes due it publishes on its own, internally and to the configured CMS, with no further call. Pass publish_at null to unschedule (the article goes back to draft). Get the dates from suggest_publishing_slots rather than inventing them - it returns instants in the project timezone that avoid days already taken. Refuses an already-published article. Unlike publish_article this does NOT need a cover or a meta description yet, but the article must have both by the time the slot fires or it is parked in the review queue instead: the response says what is still missing.

Set an article's cover photo from a public URL, from a file in the asset library (asset_id), or by promoting one of the images already in the body (image_index, from list_article_images). The file is copied into the project CDN first unless rehost is false. Every article needs a cover before publish_article accepts it, so this is usually the last thing missing on a draft written from a topic with no image agent. Use generate_image instead when the picture has to be drawn.

Unpublish an article (set its status back to draft). Also refreshes the linked translations and the recommendation references that point to it. Idempotent: unpublishing a draft is a no-op.

Update core fields of an existing article in the targeted project (title, handle, md, cover, meta, categories, status, date). Only the fields you pass are changed. Renaming the handle and flipping the status also refresh the copies other articles hold, so translations and recommendations stay in sync; switching to "published" needs a cover and a meta description.

Copy an image or video from a public URL into a project's asset library, so it is served from the project's own CDN and can be used as an article cover or embedded in markdown. Use this for any picture that does not come from generate_image - a screenshot, a logo, an image the user found - because an article must not point at a file hosted somewhere else. Returns the durable URL to use. Images and videos only, 15MB max; the file name must be free in this project (pass filename to rename it). To put a photo straight into an article instead, add_article_image and set_article_cover do this copy for you.

Permanently delete an article from the targeted project. This cannot be undone.

Permanently delete one file from a project's asset library: the record AND the stored file, so every article still pointing at its URL loses the image. Take the asset_id from list_assets (or from what generate_image / upload_asset returned). The call is refused when an article uses the file, and tells you which ones: read that list to the user and only pass force true if they confirm. This cannot be undone.

List the automation routines of the targeted project: schedule, pipeline, output policy, guardrails, next/last run times, and each routine's latest run outcome. Optionally include recent run history.

Pause an automation routine (it stops firing on schedule; manual runs stay possible). Pass resume: true to re-enable it instead -- the next occurrence is recomputed from its schedule in the project timezone.

Run an automation routine immediately (out of schedule). The run obeys the routine's guardrails and output policy; it produces articles asynchronously (follow progress with list_routines include_runs, or list_jobs). Fails if a run of this routine is already in progress. Spends AI credits.