FRIDAI

Stores the exact candidate as a reviewable and reproducible test artifact. It does not change production or require the user to visit Dashboard.

create_rule_proposal

Inserts, replaces or appends passages in owned Notebook text and leaves every other character unchanged. Each anchor is existing Notebook text that must match exactly and appear only once. The SHA-256 must match the latest content, and the previous content is always saved as a version first.

patch_notebook_content

Read-only. Returns the built-in FRIDAI 優化大師 (Notebook Optimizer) workflow. No package download, local Skill, activation or optimizerRunToken is required. It covers the current three-pass analysis, shortest-equivalent Notebook compaction with a 15,000 to 20,000-character target, fact-ledger, rewrite-fidelity, contradiction-resolution, structural audit, and same-scenario two-round behavioural comparison workflow. Short complete Notebooks stay short; all facts, conditions and fixed payloads are preserved.

get_notebook_optimization_protocol

Optional legacy FRIDAI 優化大師 Skill package download, activation and receipt verification for clients that use a local adapter. The optimizer is built into this server and works without any download, executable, local Skill or token. A legacy receipt proves package freshness, not model compliance or quality.

get_notebook_optimizer_skill

Start here when the user asks 優化大師, 優化筆記本, 檢查筆記本, 筆記本健檢 or Notebook Optimizer. With empty arguments, loads the current built-in workflow and this login's eligible brands together. Selects the only eligible subscription; asks which brand when several match. Read-only startup, not a completed audit. No local Skill, package, activation, token or user-pasted diagnostic commands are needed.

start_notebook_optimizer

Returns the latest effective notebook, SHA-256 digest, linked-rule state, business information, and safe reply settings for diagnosis.

get_ai_rule_bundle

Reads an owned proposal’s status and staleness. Publishing authorization comes from the current MCP write scopes, not a per-proposal Dashboard grant.

get_rule_proposal

Returns the dashboard analytics: totals, recent post metrics, best posting times, content insights and the format and topic ranking compared with the account's usual post.

get_autopost_performance

Summarizes the AutoPost settings as shown in the dashboard: on/off, writing modes, persona, topic sources, Stories, first comment, hashtags, collaborators, smart optimization, cross-platform sync, weekly times and timezone.

get_autopost_settings

Shows which of the four AutoPost setup steps are done, the next step, and a dashboard link to it.

get_autopost_setup_progress

Returns one manageable Notebook version so it can be reviewed or restored.

get_notebook_version

For AI clients that can run commands on the user's computer: returns a 10-minute HTTP PUT upload link for one local file and the mediaUrl it gets once registered in the media library.

create_autopost_media_upload

Read-only. For one owned or full-shared subscription, returns whether the AI switch is on, why recent customer messages were skipped, whether human-handoff notifications reached LINE/Telegram, the questions the AI handed to a person together with what the person answered, owner replies that correct or apologise for the AI, what the owner personally tells customers, repeated auto-reply texts, and de-identified customer message topics. Phone numbers, emails, street addresses, ID numbers, filled-in contact fields, account handles, customer names and customer links are masked unless the same value already appears in the brand Notebook.

get_notebook_evidence

Returns the complete allowlisted persistent Notebook and Reply Settings catalog, including scopes, ranges, and platform restrictions.

get_ruleops_capabilities

Reads the keyword auto-replies (關鍵字自動回覆 / 私訊關鍵字回覆) of every account in a subscription: whether they are on, each keyword with its fixed replies, the random delay, and the sha256 needed to change them. The replies are sent as saved, without AI.

get_keyword_replies

Returns the current FRIDAI product, feature, navigation, setup, troubleshooting, registration, and evidence-boundary guide. It contains no customer data and does not provide current billing or subscription state.

get_fridai_product_knowledge

Lists recent proposals for this key owner, including review and publication status. Rotating a key does not hide prior proposals.

list_rule_proposals

Lists the AutoPost workspaces this login owns or manages: brand, connected posting accounts and whether auto-posting is on.

list_autopost_accounts

Read-only list of Message Center conversations for brands owned by this login email or shared to it with full-manager access.

list_message_center_conversations

Lists version metadata for an owned Notebook without returning customer conversations or credentials.

list_notebook_versions

Lists active RuleOps subscriptions owned by this login email or shared to it with the full-manager role. No access tokens or conversations are returned.

list_rule_accounts

Lists the workspace's scheduled multi-platform posts — the same ones as the dashboard content calendar: date and time, platforms, a short caption preview (up to 200 characters), per-platform result with the public link once published, and whether each can still be changed or cancelled. Up to 20 per page.

list_scheduled_posts

Lists the images (or videos) in this workspace's FRIDAI media library, newest first, with the URL a draft or scheduled post can use as its image or video. Up to 20 per page.

list_autopost_media

Lists upcoming or recent posts with status, platform, time, a short caption preview (up to 200 characters) and the public link once published. Up to 20 per page.

list_autopost_posts

Pauses or resumes auto-posting. Resuming checks Instagram publishing authorization first, like the dashboard switch. Preview first, then confirm with the returned token.

set_autopost_active

Creates a five-minute CSV download for owned and full-shared brands only. The download rechecks current access.

prepare_message_center_export

Returns the owned pending proposal’s baseline, candidate, synthetic scenario blueprints, and scoring rubric. Your Claude/Codex performs generation and judging with its own quota. FRIDAI calls no model, reads no conversations, stores no test text, and permits no side effects.

prepare_rule_reply_test

Publishes the exact non-stale proposal when the current MCP connection has every required write scope. No per-proposal Dashboard grant is required. Ownership, current state and automatic backup are revalidated atomically.

publish_rule_proposal

Returns one authorized conversation with read-only delivery diagnostics: exact inbound-event generation decisions, checkpoint reasons and current pause state when the physical account is unambiguous. Current pause state is not proof of a historical cause; completed generation is not a delivery receipt. It cannot send, edit, delete, retry, or publish messages.

get_message_center_conversation

Restores one owned Notebook version after confirming the latest SHA-256. The current text is backed up before restoration.

restore_notebook_version

Saves the user's own text, optionally with an image or video already in the account's media library, for a date and time. By default it is a draft that is never published until approved in the dashboard; with autoPublish: true it is scheduled and published automatically at that time once the user confirms the preview. Preview first, then confirm with the returned token.

create_autopost_draft

Schedules 1–30 posts at once, like the dashboard composer: each post has a date and time, the platforms (Instagram, Facebook, Threads, X, LinkedIn, YouTube), shared text, optional media from the media library (image, 2–10 image carousel or video), hashtags, a first comment, a video cover, an Instagram Trial Reel option, 留言領取 (commentToDm: people who comment a keyword get a public reply and a private message; Instagram, Facebook, Threads), and per-platform text, media or account. The first call checks every post and returns one preview of the whole batch plus a confirmationToken (or the problems to fix); a second call with confirm: true and the token saves all posts together, and each is published automatically at its time. TikTok is scheduled in the dashboard.

schedule_posts

Proposes five short-video topics drawn from the account's AI rules (notebook), after checking the last 7 days of news for its field. Each topic comes with the script structure, audience, takeaway and call to action that fit it, ready to use as the brief for a spoken short-video script. Nothing is saved or published and the posting quota is not used.

suggest_autopost_video_topics

Directly replaces owned Notebook text only after the supplied SHA-256 matches the latest content. The previous content is always saved as a version first.

update_notebook_content

Directly applies allowlisted account and subscription Reply Settings. Touched previous values are saved, ownership is revalidated, and stale or duplicate writes fail closed.

update_reply_settings

Saves the keyword auto-replies (關鍵字自動回覆 / 私訊關鍵字回覆) of one account: switch them on or off, replace the keyword rules, or set the random delay before a reply. rules replaces every editable rule of the account, so include the rules to keep; rules managed by FRIDAI (editable: false) are kept automatically. baseSha256 is the sha256 last read for that account, and a stale one is refused. The previous values are returned so the change can be undone.

update_keyword_replies

Directly updates AI active state, structured business information, or an owned Notebook link. Touched previous values are saved and stale-checked before the atomic update.

update_notebook_settings

Writes text only, in the account's saved voice: a post draft for its platform (saved as a draft awaiting approval, never published automatically) or a spoken short-video script returned as text. For a script, pass structure (one of eight spoken-script structures) so it follows every part in order; length is short, standard (default, about 600 Chinese characters) or long; audience, takeaway and callToAction sharpen it; recent news is checked first unless checkRecentNews is false. Uses the normal AutoPost quota. Preview first, then confirm with the returned token.

generate_autopost_text_draft