Official Markaestro plugin, Agent Skill, and MCP server for AI agents (Claude, ChatGPT, Cursor, and more)
FAQ (answers come from our records)
How do I install it?
We did not find an install command for this repository (our install coverage is only about 4%). Check its README or releases page.
Who is it for?
We filed it under these tags: MCP 服务. The tags come from the entry itself, not a taxonomy we imposed.
Who maintains it?
repository markaestro/markaestro-agents;first catalogued 2026-10-09;last seen 2026-10-16。There is no official certification or quality rating here — we record sources and dates only.
Markaestro for AI agents
以下摘录自该仓库的 README(片段) · 在 GitHub 读完整 README →
Official agent tooling for Markaestro, the social
publishing workspace. Agents schedule, publish, and review posts on Facebook,
Instagram, TikTok, Threads, Pinterest, LinkedIn, and X, read brand and
per-post analytics, and manage Intelligent Evergreen queues.
| Folder | What it is |
| --- | --- |
| plugin/ | Claude Code plugin: the skill plus the hosted MCP server |
| skills/markaestro/ | The markaestro Agent Skill, usable by any client that supports skills |
| mcp/ | @markaestro/mcp, the MCP server as a local stdio package |
Connect
The hosted MCP server needs nothing installed and no key pasted. Add
https://markaestro.com/api/public/v1/mcp to your client; the first tool
call opens the browser to sign in, pick a workspace and brand, and click
Allow (OAuth 2.1 with PKCE and dynamic client registration).
**Claude Code**
**Claude (claude.ai and Claude Desktop):** open Customize, Connectors, click
Add custom connector, paste https://markaestro.com/api/public/v1/mcp, leave
the OAuth client fields empty, and click Add. Click Connect to sign in.
**Cursor, ChatGPT, Grok, OpenClaw, Hermes, and other clients:** step by step
instructions for each are at
markaestro.com/developers/agents.
**Skill only** (Claude Code, Cursor, Codex, Copilot, Gemini, and other agents
that read skills; listed on skills.sh):
**OpenClaw** (listed on ClawHub):
**Local package** (reads media from your own disk; needs a workspace API key
from Settings, API Access):
Example prompts
- "What did we post on Instagram last month, and which three posts got the
most engagement?"
- "Draft a LinkedIn post announcing our new cold brew and schedule it for
Tuesday at 9am New York time. Don't publish anything else."
- "When does our audience respond best? Put next week's three drafts in
those slots."
Safety
- The user scopes every connection at sign-in: one brand, or all brands in
the workspace. A single-brand connection cannot reach any other brand; an
all-brands connection names the brand on each post it creates.
- Agents manage social media, not the account: no tool reaches account
settings, billing, team members, API keys, webhooks, or channel
connections, and none deletes a published post, takes one down from a
platform, or archives an Evergreen queue. The key issued at the agent
sign-in carries the same limits at the REST layer, so they hold whichever
client holds the token.
- create_post saves a draft unless scheduledAt is set. publish_post is
the only tool that publishes immediately; scheduling (scheduledAt,
bulk_posts, activating an Evergreen queue) sets up future publishes. The
skill tells the agent to ask the user before publish_post and before
activating an Evergreen queue.
- Every tool declares readOnlyHint, destructiveHint, and openWorldHint
explicitly, set from what it does: destructiveHint on tools that edit,
remove, unschedule, or publish (update_post, delete_post, bulk_posts,