← MarkaestroCONTENT HISTORY
WHAT CHANGED
Update to Markaestro
Snapshot Nov 18, 2026 · 06:18 UTC · version 0.4.0
Collection source: downloaded plugin package.
First saved snapshot
No earlier snapshot is available to establish a change.
Save Markaestro and check future changes in Following.
Up to 3 plugins free · no card required.
Compare saved observations
Download comparison JSONFull technical diff · 0 changed fields
Full snapshot data
{
"description": "Schedule, publish, and review social posts through Markaestro (Facebook, Instagram, TikTok, Threads, Pinterest, LinkedIn, X, YouTube), read brand and per-post analytics, and manage Intelligent Evergreen queues using its MCP server or public API. Use when a task mentions Markaestro, posting or scheduling to a connected social account, uploading media for a post, checking whether a post published or how it performed, reporting on social performance, deciding what or when to post next, batch-scheduling or editing scheduled content, refreshing analytics, finding the best time to post, writing in a brand's voice, or managing proven recurring content.",
"included_files": [
{
"relative_path": "references/settings.md",
"size_in_bytes": 3375
},
{
"relative_path": "references/tools.md",
"size_in_bytes": 8430
}
],
"name": "markaestro",
"skill_md_contents": "---\nname: markaestro\ndescription: Schedule, publish, and review social posts through Markaestro (Facebook, Instagram, TikTok, Threads, Pinterest, LinkedIn, X, YouTube), read brand and per-post analytics, and manage Intelligent Evergreen queues using its MCP server or public API. Use when a task mentions Markaestro, posting or scheduling to a connected social account, uploading media for a post, checking whether a post published or how it performed, reporting on social performance, deciding what or when to post next, batch-scheduling or editing scheduled content, refreshing analytics, finding the best time to post, writing in a brand's voice, or managing proven recurring content.\n---\n\n# Working with Markaestro\n\nMarkaestro is a social publishing workspace. Agents reach it through the\nMarkaestro MCP server (preferred) or the public API at `/api/public/v1`.\nThe hosted MCP server signs the user in through the browser (OAuth) and\nreceives a key scoped at consent to one brand or to every brand in the\nworkspace; the REST API takes that same kind of key as a bearer token.\n\nTool-by-tool inputs and example outputs are in\n[references/tools.md](references/tools.md). Delivery modes, post statuses,\nand platform settings are in [references/settings.md](references/settings.md).\nRead those when you need an exact field name.\n\n## Before anything else\n\n1. Confirm the MCP server is connected: a `list_products` tool should be\n available. If it is not, the fix depends on the client you are running\n in. Never ask the user for an API key when the client can open a browser;\n the sign-in hands the client a key scoped to the brands the user picks.\n - Installed but not signed in (a 401 or \"needs authentication\" state):\n Claude Code: run `/mcp`, pick `markaestro`, finish the sign-in.\n Cursor: Cursor Settings, Tools & MCP, click Markaestro (Needs login).\n ChatGPT: reconnect the Markaestro app in Apps & Connectors.\n Grok Build: the next tool call reopens the sign-in; grok.com: reconnect\n the connector. OpenClaw: `openclaw mcp login markaestro`, then\n `openclaw mcp reload`. Hermes: `/reload-mcp`, then call a tool.\n - Not installed at all: send the user to\n `https://markaestro.com/developers/agents?client=<claude-code|claude|cursor|chatgpt|grok|grok-bot|openclaw|hermes>`,\n which opens that client's steps. For Claude Code the whole setup is\n `claude plugin marketplace add markaestro/markaestro-agents` then\n `claude plugin install markaestro@markaestro`; any client can also add\n the server URL `https://markaestro.com/api/public/v1/mcp` directly.\n - Headless or CI only (no browser), and Grok Bot: a workspace API key from\n Settings, API is passed as a header instead:\n `claude mcp add --transport http markaestro https://markaestro.com/api/public/v1/mcp --header \"Authorization: Bearer mk_...\"`,\n or the local package: `claude mcp add markaestro -e MARKAESTRO_API_KEY=mk_... -- npx -y @markaestro/mcp`.\n `https://markaestro.com/developers/agents?client=headless` has the same\n header for other clients' config files.\n2. Call `get_channel_rules` once per session. It returns the per-channel\n rules and `keyMode`. If `keyMode` is `test`, say so when reporting\n results: test keys never reach a real platform.\n3. Call `list_products`. The brands in the answer are the ones this\n connection can act on: one for a single-brand connection, every brand\n in the workspace for an all-brands one.\n - Several brands: pass `productId` to `create_post`, `create_posts`,\n `list_posts`, and `create_evergreen_queue`, and to `get_analytics` or\n `list_post_analytics` to report on one brand (without it they cover\n the whole workspace). Confirm the brand with the user when the\n request does not name one.\n - One brand: every call uses it; `productId` can be left out.\n - The user names a brand that is not listed: they reconnect from their\n client (Claude Code: `/mcp`, sign out, sign in) and pick that brand,\n or All brands, at consent.\n\n## The posting model\n\n- `create_post` **saves a draft** unless `scheduledAt` is set. A draft never\n reaches a platform on its own.\n- `scheduledAt` (ISO 8601, UTC) makes the post `scheduled`; the worker\n publishes it at that time. Convert the user's local time to UTC and echo the\n UTC time back to them.\n- `publish_post` publishes **now**. Ask the user before calling it for anything\n public. It returns a job run; poll `get_job_run` until `succeeded` or\n `failed` and report the message.\n- Facebook, Instagram, and TikTok require an explicit `deliveryMode` when\n scheduling: `direct_publish` to post through the platform API, or\n `manual_reminder` for a timed reminder the user posts by hand. Threads,\n Pinterest, LinkedIn, X, and YouTube default to `direct_publish`.\n- One post, several channels: pass `targets` (one entry per channel) instead\n of `channel`. Each target can carry its own `destinationId` and\n `deliveryMode`. Platform `settings` may ride on one target only.\n- A week of content: `create_posts` takes up to 25 items in one call and\n reports success or failure per item. Check every `ok` before summarizing.\n- To change a draft or scheduled post, `update_post` edits its caption,\n media, one channel's `settings`, or (scheduled only) its time; omitted\n fields stay as they are. Published and failed posts cannot be edited, and\n channels are fixed once a post exists.\n- Before writing captions, read `get_brand_profile {productId}` for the\n brand's voice and identity. Before picking a time, `suggest_post_times`\n gives the brand's best windows, learned from its own history (null until\n there is enough; then fall back to asking the user).\n- TikTok Direct Post: call `get_tiktok_posting_options` first and use one of\n its `privacyLevelOptions` in the tiktok `settings`.\n\n## What agents cannot do\n\nAccount settings, billing, team members, API keys, and connecting or\ndisconnecting social accounts are not available to agents; send the user to\nMarkaestro for those. Agents also cannot delete a published post or take one\ndown from a platform, and cannot archive an Evergreen queue: say so and\npoint the user to Markaestro rather than looking for another way.\n\n## Media\n\n- Upload first with `upload_media` (a public URL or data URL; a file path\n only with the local `@markaestro/mcp` package, since the hosted server\n cannot read your disk), then pass the returned asset id in\n `mediaAssetIds`. Uploads count against the monthly quota, so reuse an asset\n from `list_media` when the same file was uploaded before.\n- Instagram, TikTok, and Pinterest require media. LinkedIn requires text.\n Video must be the only media item on LinkedIn and Pinterest. YouTube takes\n exactly one video; a vertical or square video of 3 minutes or less becomes\n a Short on YouTube's side.\n\n## A safe scheduling flow\n\n```\nget_channel_rules\nlist_products -> productId, connected channels\nlist_destinations {productId} -> only if a channel has several accounts\nupload_media {source} -> asset id (skip for text-only)\ncreate_post {caption, targets|channel, mediaAssetIds, scheduledAt, deliveryMode}\nget_post {postId} -> confirm status and scheduledAt\n```\n\nReport back the post id, status, channels, and the UTC schedule. Mention that\nit can be edited with `update_post` or cancelled with `delete_post` until it\npublishes.\n\n## Reading the schedule\n\n- `list_posts {status: \"scheduled\"}` is the queue; `published`, `failed`,\n `platform_action_required`, and `draft` cover the rest.\n- `platform_action_required` means a person must finish the post: TikTok inbox\n handoff, or a manual reminder. Say what the next action is; do not treat it\n as failed. Once the user confirms they posted it, `mark_post_posted` (with\n the live link if they have it) records it as published.\n- `partial_failed` means some targets published and some did not. Read\n `publishResults` on the post before retrying anything.\n\n## Reading performance\n\n- `get_analytics` is the brand overview for a window (`days`, or `since`\n and `until`): totals with the prior period, per-channel rollups, daily\n series, engagement breakdown, follower trend, top posts, a weekday-by-hour\n heatmap, content-type averages, and `insights` (plain-language findings,\n each with its `sampleSize`). Read it before recommending what, when, or\n where to post, and quote the sample size when you repeat an insight.\n- `list_post_analytics` is one row per post with its latest metrics,\n including posts published directly on the platform (`source: \"native\"`)\n that the connected account was scanned for. `sort: \"engagements\"` or\n `\"views\"` finds what worked; `get_post` has the full caption and media\n for a Markaestro row, and `externalUrl` opens a native one. Pass\n `source` to compare the two halves of the account.\n- Numbers look stale, or the user just posted: `refresh_analytics` pulls\n live metrics from the platforms now. It is limited to a few calls a\n minute; when `remaining` is above zero, a second call continues the window.\n- To take a published post down, the user does it in Markaestro or in the\n platform's own app; agents cannot. `delete_post` only removes drafts and\n posts that have not reached a platform.\n- `get_post_analytics_history` shows how one post earned its numbers (the\n 1h to 90d snapshots with growth between them) and whether polling is still\n `active`. A post under 24 hours old is not yet comparable to older ones.\n- `coverage.bySource` in `get_analytics` says how many analyzed posts were\n Markaestro posts and how many were native. A native post discovered\n today has a `discovered` snapshot only; its history fills in from there.\n- The window is clamped to the plan's history; `window.maxDays` in the\n answer says the cap (`-1` is unlimited). If the user asks for more than\n the plan keeps, say so rather than presenting the clamped window as the\n full one.\n- `null` means the platform does not report that metric (TikTok and Threads\n never report reach). Never read it as zero, and use\n `engagementRateByViews` where reach is missing.\n- These tools need the `analytics.read` scope. A 403 here means the key\n predates the scope: the user creates a new key, or reconnects the agent.\n\n## Intelligent Evergreen\n\n- Call `preview_evergreen_queue` with a published post id first. It returns\n measured eligibility and a recommended cadence without writing anything.\n- `create_evergreen_queue` creates a draft policy. It does not schedule posts.\n- Ask the user to confirm the cadence, channels, variants, and review policy\n before calling `activate_evergreen_queue`. Show every caption variant and,\n once the user confirms they are accurate, pass `contentConfirmed: true` on\n `create_evergreen_queue` or `update_evergreen_queue`. Activation and resuming\n answer `EVERGREEN_CONTENT_REVIEW_REQUIRED` until that confirmation exists.\n- `review_each_run` creates a draft occurrence for each run. It never\n auto-publishes. `approve_future_runs` creates ordinary scheduled posts.\n- Pause stops the queue and unschedules its pending occurrence; resume\n restarts it. Archiving is permanent and stays with the user in Markaestro. Use `list_evergreen_runs` to report evaluation outcomes and\n `get_evergreen_analytics` for lifetime metrics and attributed conversions.\n\n## Errors\n\nEvery failure carries a stable `code`. The common ones:\n\n| Code | Meaning | What to do |\n| --- | --- | --- |\n| `UNAUTHENTICATED` | Key missing, revoked, expired, or wrong mode | Ask the user for a valid key |\n| `FORBIDDEN` | Key lacks the scope | Ask for a key with the needed scope |\n| `VALIDATION_ERROR` with `issues[]` | One or more targets invalid | Fix each listed channel and retry |\n| `VALIDATION_SCHEDULED_DELIVERY_MODE_REQUIRED` | Scheduled Meta or TikTok post without `deliveryMode` | Add `deliveryMode` |\n| `VALIDATION_POST_IS_PUBLISHING` | Post mid-publish | Wait, then retry |\n| `RATE_LIMITED`, `RATE_LIMITED_CHANNEL` | Too many calls or channel ceiling | Wait for `Retry-After`; the server already retried once |\n| `QUOTA_EXCEEDED_*` | Plan limit reached | Tell the user which limit |\n| `NOT_FOUND` on a post | Not this brand's post | Check the id and the key's brand |\n\nNever invent a post id, an asset id, or a destination id. Read them from the\ntools.\n\n## Without the MCP server\n\nThe same operations exist as REST calls with\n`Authorization: Bearer mk_...` against `https://markaestro.com/api/public/v1`.\nThe OpenAPI description is served at `/api/public/v1/openapi.json`. Send an\n`Idempotency-Key` header on every mutation.\n"
}SHA-256 of public snapshot: 58e0bc30522241c0be7d3e20411220c9da0632cf0734bf87856f9cf88e176128