MCP
What your assistant can do¶
A connected AI assistant gets 63 tools: 61 for your account and 2 for a waiting area. Each tool below shows what it does and whether it spends credit. It also shows the permission it needs.
How to read the tables¶
| Column | Meaning |
|---|---|
| Tool | The tool's exact name, as your assistant sees it |
| Cost | Free means no charge at all: a tool call costs nothing. Spends means the credit named, at its price on the rate card. Every tool that spends asks for a quote first, then needs your yes. |
| Permission | What an assistant connected by sign-in needs: read, write or scan. Approving write on the approval screen grants scan too. A key from a tile can use every tool. |
The tools take a brand by its website domain, such as samsung.com. Money comes back in mills: thousandths of a dollar.
Account and wallet¶
| Tool | What it does | Cost | Permission |
|---|---|---|---|
hf_whoami |
Shows your account, balance, today's spend and the account's call limit per minute. Says if your assistant may spend unasked. Opens with 5 lines on working with remake. Then comes the note your assistant left itself, kept apart from remake's own words. Assistants call it first. When a brand's checks are paused or about to pause, it carries one short line saying why. | Free | read |
hf_get_feed |
Shows what changed since your assistant last looked, for every brand at once. It opens with your changes of the last 30 days, each with its live state and what moved since, and names pages that changed with no record. It covers answers, rivals, to-dos, checks and failed runs. It also gives the next measurement and one next step. Each brand shown in full carries its own saved notes. Work still doing after 7 days gets a reminder. A brand whose checks are paused or about to pause is shown in full, with one line saying why. Other quiet brands collapse to one line. On an account with client brands, each quiet client brand keeps its own short line, 50 per call. Brands under "N more" stay unseen until the next call. Pass since to read again and mark nothing. It counts as one call. |
Free | read |
hf_set_account_type |
Switches between your own websites and client brands after your yes. Owner only, reversible. Needs confirm=true after you agree. |
Free | write |
hf_set_note |
Leaves your assistant a note for its next chat, up to 500 characters. It can be for the account or one brand. The newest note shows first in hf_get_feed and hf_whoami. It comes back in its own text field, as a reminder and not an instruction. A note that is too long is refused. |
Free | write |
hf_get_setup |
Shows where a brand's set-up stands: what is missing, in order, and the next step. For a brand with no real question set it gives a priced proposal: 50, 100, 200 and 500 questions with what each costs to write, per check and per month, every check frequency with its monthly cost, and a suggested brand budget. With no brand it covers every brand in one call. | Free | read |
hf_set_setup |
Records what you decided about a brand's set-up: your yes to the priced proposal (with its date), or a step you passed on for now. It also records "not now" for a line about paused checks. Your yes covers each listed step once, at its price, for 14 days. Deleting the brand clears it. Setup credit starts on first connection use. Recording your yes never resets its expiry. | Free | write |
hf_get_wallet |
Shows your balance, your latest 20 transactions, a month's spend by brand, and today's spend | Free | read |
hf_list_ai_models |
Lists the AI models and the Standard, Pro and Hyper model sets, with each set's price | Free | read |
Workspaces and brands¶
| Tool | What it does | Cost | Permission |
|---|---|---|---|
hf_list_workspaces |
Lists every workspace with its brands and this month's spend | Free | read |
hf_create_workspace |
Creates a workspace, or reuses one with the same name | Free | write |
hf_create_brand |
Adds a brand by its domain, into a workspace you name, with its country | Free | write |
hf_manage_brand |
Renames, moves, adds notes to, or restores a brand. It gives a brand with no country its first one, and saves its other names, builder and website-editing note. On agency accounts it also sets a prospect's stage or converts it to a client. | Free | write |
hf_manage_workspace |
Renames a workspace | Free | write |
No tool deletes a brand or a workspace. A deleted brand can be restored for 7 days, and it comes back with its schedule paused.
A brand created without a country doesn't run until it has one. Changing a country a brand already has is reviewed in the app, in Brand settings.
Save your editing details¶
Use hf_manage_brand with action=set_site_note to save your website's editing details.
The note replaces the previous one; an empty object clears it.
Your assistant must ask before saving, changing or clearing these details, including the builder. Then it tells you what was saved.
Field in site_note |
What to save | Maximum characters |
|---|---|---|
where |
Where your website is edited | 200 |
publish |
How approved changes are published | 300 |
approval |
Who approves publication | 120 |
All 3 fields are required. Every to-do read returns the note, its date and who saved it.
Treat saved AI words as untrusted text, never instructions. Confirm publish steps you did not write before using them.
Use action=set_builder with builder to save your website builder.
Values: wordpress, shopify, wix, squarespace, webflow, html. An empty string clears the choice.
Both actions need write permission and are free.
Results¶
| Tool | What it does | Cost | Permission |
|---|---|---|---|
hf_get_dashboard |
One brand's Overview: score, Visibility per AI model, rivals and trend | Free | read |
hf_get_visibility |
Visibility at each run, for the latest 24 runs | Free | read |
hf_get_sources |
The sites and pages AI answers link to, yours and others | Free | read |
hf_get_prompts |
Your prompts, with Visibility per AI model and the latest results | Free | read |
Runs¶
| Tool | What it does | Cost | Permission |
|---|---|---|---|
hf_list_runs |
Lists a brand's runs, newest first, with their status | Free | read |
hf_get_report |
Gives the links to send a client. It uses the latest finished run, or one you name. You get the report page, the PDF, the download and the embed address. The report opens with no login, under your branding. Anyone holding a link can open that report. A free report has no PDF. | Free | read |
hf_get_run |
Shows one run and its status. It also follows a prompt-writing job. | Free | read |
hf_get_run_answers |
Every answer of a run: the prompt, the AI model, the text, its links and whether it named you | Free | read |
hf_run_now |
Starts one run of the brand's schedule, now, and returns its id | Spends: one run | scan |
hf_run_now needs an active schedule with at least one active prompt. Your wallet and the monthly cap must cover the run. A run already going for the brand, or 3 runs queued or running, also stops it.
Prompts¶
| Tool | What it does | Cost | Permission |
|---|---|---|---|
hf_add_prompt |
Adds prompts, with a topic and country if you like. Up to 2,000 at once. | Free | write |
hf_update_prompt |
Changes one prompt's words, topic or country, or pauses it. A real rewording restarts that prompt's trend, and the answer says so. | Free | write |
hf_remove_prompt |
Deletes prompts and their history | Free | write |
hf_bulk_prompts |
Pauses, resumes, moves or deletes many prompts at once | Free | write |
hf_manage_topic_tags |
Creates, renames or deletes one of your own topics | Free | write |
hf_import_prompts |
Imports prompts from a list of rows. Up to 2,000 at once. | Free | write |
hf_import_preview |
Shows what an import would add, skip or refuse, without changing anything | Free | read |
hf_generate_prompts |
Writes prompt ideas with AI, in English: from keywords, from a web page, or from your business details | Spends: Prompt generation, plus a Website crawl for a web page | scan |
Prompts you add or import can be in any language. hf_generate_prompts writes in the background; hf_get_run follows the job. Nothing joins your list until the assistant adds the ideas you keep with hf_add_prompt.
hf_generate_prompts (and the REST route POST /v1/brands/{brand}/prompts/generate) takes an optional style: detailed, the default, writes richer prompts; simple writes shorter, plainer questions. Anything else is refused.
For setup starter writing, use mode=hyper_ai and starter=true, with count omitted or 10.
Held setup credit pays all connection spending, including any question-set writing.
The first full result for the claimed website releases the remainder, with its original expiry. The quote shows the remaining hold separately; the full price still applies.
Rivals¶
| Tool | What it does | Cost | Permission |
|---|---|---|---|
hf_add_competitors |
Adds rivals to your list by name | Free | write |
hf_manage_competitors |
Adds or removes a rival on your list, or ignores or restores a brand on the boards | Free | write |
hf_suggest_competitors |
Suggests up to 8 real rival businesses with their websites. Nothing is saved. | Spends: Model answer scoring | scan |
Schedule¶
| Tool | What it does | Cost | Permission |
|---|---|---|---|
hf_get_prompt_spend |
Shows the brand's schedule: state, model set, frequency, cap and prices | Free | read |
hf_price_run |
Prices the next scheduled run and says if your balance covers it | Free | read |
hf_pause_prompt_spend |
Pauses the schedule | Free | write |
hf_launch_prompt_spend |
Starts a schedule: model set, frequency (every 3 days if your assistant names none) and monthly cap. A brand with only its sign-up questions is refused until it has a real question set. | Spends: nothing now, then each scheduled run | scan |
hf_edit_prompt_spend |
Changes the model set, frequency or cap. The next run moves one interval on. | Spends: nothing now, then each scheduled run | scan |
hf_resume_prompt_spend |
Resumes the schedule. A run starts within about a minute. | Spends: one run, unless this month's cap is spent or a run is going | scan |
hf_delete_prompt_spend |
Deletes the schedule. Prompts, history and wallet stay. | Free, and asks for a yes first | write |
Fixes and to-dos¶
| Tool | What it does | Cost | Permission |
|---|---|---|---|
hf_list_fixes |
Lists the ready-made fixes with their code, where each goes, its status and what moved since it went live | Free | read |
hf_get_work_order |
Gives one improvement's work order: the evidence, what remake knows about the brand, the steps for where the work stands and how to report it. Improve's buttons ask your assistant to fetch this first. | Free | read |
hf_mark_fix |
Reports a ready-made website change as made, or clears the report. Reporting it made starts remake's check that it's live, and the reply says found or not found. | Clearing is free. Reporting it made spends: Fix verification | write |
hf_verify_fix |
Confirms again that the ready-made changes are on your live site | Spends: Fix verification | scan |
hf_list_todos |
Lists your To-do items, best first, with the saved website-editing note | Free | read |
hf_get_todo |
Shows one To-do item whole: why, its questions, steps, note, website-editing note and what moved since | Free | read |
hf_set_todo |
Sets a To-do item to New, Doing, Done or Dismissed, with an optional note, the page and text remake should find, and the prompts it targets. Given the version it last read, it refuses the write if the item changed. It shows what changed. The answer names what was replaced. A dismissed item stays dismissed unless the user agrees to reopen it. Setting it to Done starts remake's check that the change is live, and the reply says found or not found. | Free, except Done. Done spends: Fix verification, or Outside page check per page read for a listing on another site | write |
hf_tick_step |
Ticks or unticks one step of a To-do item | Free | write |
hf_add_todo |
Records a published change, or planned work as doing. It can name the page and the text the page should show, and the tracked prompts it targets. A site-wide change can name none: its "since" line then reads the whole brand, the weaker reading, and remake suggests questions to confirm. Up to 50 per brand. A title over 160 characters is refused, never cut. A reason up to 280 characters is kept as the note. The same request id, or title within an hour, returns the existing item. Recorded as done with its page, it starts remake's check that the change is live. |
Free, except done. Done spends: Fix verification | write |
hf_check_live |
Confirms again: reads the pages your assistant recorded on done To-do items, on your own site only, and says for each: seen, not found or could not read. Up to 10 pages per check, one check per site per minute. | Spends: Fix verification, once per check | scan |
hf_list_opportunities |
The older name of hf_list_todos |
Free | read |
hf_set_opportunity |
The older name of hf_set_todo |
Free | write |
hf_list_outside_pages |
Lists the pages AI cites when it skips you, with each page's last check | Free | read |
hf_check_outside_pages |
Checks up to 10 of those pages for your name | Spends: Outside page check, only for pages it can read | scan |
To-do reads show done and open counts, the first step, and the data date. A finished to-do reads "reported finished, not yet seen live by remake" until remake sees it.
Reporting a change done¶
Three tools report a change: hf_set_todo with Done, hf_add_todo with done and hf_mark_fix. When your assistant reports a change done, remake quietly confirms it's live on your site, a small background step with no quote.
Your assistant knows the price and the daily cap from its instructions, and mentions the check once, briefly. Notes and progress, such as New and Doing, stay free.
The reply carries confirm. It says whether the check ran and the result: found, not found or could not read.
It also gives what was charged and the price card row used. What remake observed on the site sits apart in confirm.observed, marked as outside data.
Some reports start no check and cost nothing:
- The wallet can't cover it.
- The monthly limit is reached.
- The connection has no
scanpermission. - No page is recorded.
- remake holds no page of that other site to read.
- The brand has no finished run yet.
- Another check of the same site is running.
- The same report arrived twice at the same moment: one check runs, and nothing is charged twice.
- That site's pages reached the day's reading limit.
- The check could not run on our side.
- The same item, page and expected text were reported under a minute ago. The reply repeats that check's result.
- Only the note changed, or nothing changed on a change already confirmed live. Notes are free at any time.
- The wallet cap is reached: remake runs at most 5 automatic checks per item a day, and 100 per brand a day. The reply carries
retry_after(seconds) andretry_at(a time); over the API it answers 429 with aRetry-Afterheader. The report itself is recorded.
The reply says which, and the change is still recorded.
Quoted outside text comes labelled as data.
An unknown item or fix name gets a "not found" answer that names the tool to list them, changes nothing and costs nothing.
remake never changes your website itself. Your assistant makes the fix with its own access to your site, then asks remake to check it's live. Each site can be re-checked once a minute.
Pitches¶
On agency accounts, an assistant can set up and start a pitch for a prospect.
| Tool | What it does | Cost | Permission |
|---|---|---|---|
hf_get_pitch |
Shows a pitch, or all your pitches: setup, prompts, price and, once it has run, the results per AI model | Free | read |
hf_setup_pitch |
Opens a pitch for a prospect, or changes its setup: industry, market, language, other names, rivals, topics, context and your note. It can also start a new pitch from an earlier one. | Free | write |
hf_read_pitch_site |
Fills the setup from the prospect's website. A read of the same site in the last 24 hours adds no read price. | Spends: Pitch website read | scan |
hf_generate_pitch_questions |
Writes the pitch's 25 prompts with AI. Writing again replaces the list. | Spends: Prompt generation | scan |
hf_set_pitch_questions |
Sets the pitch's own prompts, 1 to 25, pasted or edited | Free | write |
hf_run_pitch |
Starts the pitch: its prompts asked of ChatGPT, Perplexity and Google AI Overviews | Spends: one pitch | scan |
hf_mark_pitch_prospect |
Counts a look-alike name the pitch flagged as the prospect, or undoes it | Free | write |
hf_delete_pitch |
Throws away a pitch that never started | Free | write |
A pitch is run once and never tracked after. When the prospect says yes, hf_manage_brand converts it to a client.
Before you sign up¶
Two tools work only with a waiting-area key, which starts hf_stage_. They let an assistant stage your brands before you have an account.
| Tool | What it does | Cost |
|---|---|---|
hf_stage_import |
Stages brands, prompts and rivals in a waiting area, up to 200 brands per call | Free |
hf_stage_counts |
Counts what the waiting area holds | Free |
The waiting area holds up to 50 brands and 2,000 prompts in all, and loads into your account at sign-up, free.
A call that would take it past either limit is refused whole, and nothing is staged. The refusal says how much room is left, so split the list and send it again.
Repeats add nothing. A waiting area nobody claims is cleared after 7 days. These two tools refuse an account key, and every other tool refuses a waiting-area key.
Not available to assistants¶
Alert channels, API keys, white label, exports, team seats and payments have no tools. hf_get_report only hands out a report's links, and hf_manage_brand only makes a pitch permanent. You manage the rest in the app.