iHatePosting

[iHatePosting]

ihateposting.com

iHatePosting for AI agents

Let Claude, Claude Code, Cursor, Gemini CLI, Grok Build, Kimi Code or Qwen Code draft, check, schedule and publish posts on 14 social networks, using the accounts you have connected at ihateposting.com. This package gives each agent the hosted iHatePosting MCP server (16 tools) and a skill that teaches it to post safely. Every agent here signs in with your iHatePosting account; none of them needs an API key.

GitHub Copilot in VS Code uses the same hosted server: Install in VS Code, then sign in with your iHatePosting account in the browser. There is no API key to paste. The steps are under Install.

Set-up pages for every agent: Claude · Claude Code · Claude Cowork · ChatGPT · Codex · Cursor · Gemini CLI · Grok Build · VS Code and GitHub Copilot · OpenClaw · Hermes Agent · DeepSeek Harness · Perplexity Computer · Muse · NanoClaw · Paperclip · Antigravity CLI · Amp · Cline · Kilo Code · Zed · Warp · Devin Desktop · OpenHands CLI · Freebuff CLI · omp · all of them

Some of those pages add the server by hand, with an API key; for Claude, Claude Code, Cursor, Gemini CLI, Grok Build, Kimi Code and Qwen Code, the plugin in this repository signs in instead (Install).

[License: MIT] [npm: ihateposting-mcp] [npm: ihateposting]

[M8ven: continuously verified]

Nothing goes out unless you say so. A new post is saved as a draft unless you ask for it to be published now or at a set time.

What your agent can do

• Draft, schedule or publish one post to several networks at once, either to every account you have on a network or to one account you pick. A time such as "9:00 AM" is read in the timezone set in your iHatePosting settings (Settings → Scheduling), not your computer's.
• Check a post before it exists. validate_post runs the checks the publisher runs just before it posts, and creates nothing. It also tells you when a network has no connected account, or one that needs reconnecting.
• Learn each network's rules: character limit, media rules and the options a network requires, such as a Pinterest board.
• Write per network: different text for one network, and that network's own settings. That covers threads on X, Bluesky, Threads and Mastodon, a first comment on Instagram, LinkedIn and Threads, Instagram Reels, Stories and carousels, Facebook Reels and Stories, TikTok photo posts and privacy, and a YouTube video's title, privacy and playlist.
• Attach media: upload an image or a short video with alt text, or reuse anything already in your media library.
• Manage what is already there: see what happened on each network, rewrite a draft, scheduled or failed post, move it to another time, resend only what failed, or delete it.
• Report results: per-channel numbers, follower counts, posting times and top posts, from 7 days back to everything recorded.

Every plan includes the API key that the MCP server, the REST API and the CLI use. Webhooks come with every plan too.

Tools

The hosted server offers these 16 tools, plus one the upload box uses by itself. The last column is what a tool does to your posts and accounts.

Networks

All 14 networks work through the same tools. Name one in platforms (bluesky, x, linkedin, facebook, threads, mastodon, telegram, discord, tumblr, slack, instagram, pinterest, tiktok, youtube), or pass an account id from list_accounts.

If you leave them out, create_post fills a YouTube title and a Pinterest pin title from the first line of your text, a Pinterest link from the first URL in it, and YouTube and Tumblr tags from its hashtags. validate_post does not fill them in, so it reports a missing YouTube title as a problem.

What each network accepts, with its main options, is in docs/platforms.md. The option keys a post is likely to need, with their values and defaults, are in the reference the skill loads when it needs it (skills/ihateposting/references/platform-options.md). A post can attach at most 20 media items, whatever a network allows.

Example prompts

• "What is scheduled to go out this week, and did anything fail?"
• "Draft this announcement for LinkedIn, Bluesky and X. Give X a shorter version, and check all three before you save."
• "Turn these notes into a thread for X and Threads and save it as a draft."
• "Upload this photo with alt text and draft an Instagram post with it. Put the hashtags in the first comment."
• "Which of my Pinterest boards suits this recipe? Draft a pin to it."
• "Why did yesterday's LinkedIn post fail?"
• "How did my posts do over the last 30 days, and which network did well?"

examples/ has ready-made create_post payloads for threads, Reels, carousels, YouTube, Pinterest, TikTok and more. Every one is a draft.

Install

Every agent below signs you in to your iHatePosting account in the browser. None of them needs an API key.

/plugin marketplace add iHatePostingOrg/agent-skill/plugin install ihateposting@ihateposting

{  "mcpServers": {    "ihateposting": {      "type": "http",      "url": "https://ihateposting.com/mcp/oauth"    }  }}

cursor://anysphere.cursor-deeplink/mcp/install?name=ihateposting&config=eyJ0eXBlIjoiaHR0cCIsInVybCI6Imh0dHBzOi8vaWhhdGVwb3N0aW5nLmNvbS9tY3Avb2F1dGgifQ==

{  "servers": {    "ihateposting": {      "type": "http",      "url": "https://ihateposting.com/mcp/oauth"    }  }}

gemini extensions install https://github.com/iHatePostingOrg/agent-skill

grok plugin install iHatePostingOrg/agent-skill --trust

/plugins install https://github.com/iHatePostingOrg/agent-skill

qwen extensions install iHatePostingOrg/agent-skill:ihateposting

npx skills add https://github.com/ihatepostingorg/agent-skill --skill ihateposting

Get an API key

None of the agents above needs one. A key is for the ihateposting command line, the local npm server, the REST API, and agents that cannot sign in.

Sign in at ihateposting.com, open Settings → Developers and create a key. It starts with pk_live_, and you can show it again later from the same page (a key made before keys could be shown again, in late September 2026, cannot be; the page says so, and regenerating gives you one that can). There is one key per account: regenerating switches the old one off everywhere it is used. The key can post on your behalf, so keep it out of chats, screenshots and shared files.

Safety

• Drafts by default. create_post saves a draft unless it is given action: "now" (publish at once) or action: "schedule" (publish at a set time). The CLI also saves a draft unless told otherwise. The REST API has no default at all: a request without action is refused.
• Broken posts are refused before they are saved. A post that is meant to go out is checked against every network it targets when it is created or updated, and refused with the network's reason if any of them would reject it. Drafts skip this check, so an unfinished post can be saved. The publisher checks each send again just before posting and skips one that fails.
• Option names and values are checked on every post, drafts included. A key a network does not read (madeForKids for ytMadeForKids), or a value it does not take, is refused with a message naming the right one, rather than saved and ignored.
• A draft is scheduled with update_post. reschedule_post refuses a draft. update_post with action: "schedule" runs the create-time check and splits the post per account, like create_post.
• Times come back in two forms. scheduledAt is UTC; scheduledLocal beside it is the same moment in the account owner's timezone, which whoami names as ownerTimezone. A date sent without action: "schedule", or a date that does not exist, is refused.
• A scheduled post cannot go back to being a draft. update_post refuses that. Delete the post and create it again as a draft.
• Scheduling a multi-network post splits it. When create_post or update_post schedules or publishes a post for more than one account, it becomes one post per account, sharing a group id; create_post's answer lists each one. A draft stays one post. After the split, reschedule_post and delete_post act on one of them at a time, and update_post can change one's text, options and media as long as you send that post's own accountIds — adding or swapping an account is refused.
• Deleting never unpublishes. delete_post removes iHatePosting's record only. A published post is refused unless force is set, and it stays live on the network. A post that is publishing right now cannot be deleted.
• Retry only resends what failed. A send that published is left alone. A network can report a failure for a post that went live anyway, so check the network before retrying.
• Approval prompts depend on the agent. In Claude Code the skill pre-approves only the tools that read, so anything that creates, changes, publishes or deletes asks first unless your own permission settings allow it. Cursor, Gemini CLI, Grok Build, Kimi Code and Qwen Code apply their own approval settings.
• Social accounts are connected at ihateposting.com, never through an agent. Each network's own sign-in page handles the password, and posts go out through each network's official API, the same way the iHatePosting app sends them. An agent never sees a social login.

Media

• upload_media takes either a url we fetch ourselves — preferred, and the only route that works for a video of any real size (up to 100 MB; images up to 25 MB) — or the file as base64, with an optional alt text, and returns a library id to put in mediaIds. base64 through a conversation is capped at 8 MB and is the bytes the model itself typed, so it truncates: never use it for a file that only exists as a chat attachment. open_upload_widget covers that case by showing a file picker in the conversation, and the ihateposting upload <path> CLI command covers it for an agent with a shell. Images can be JPEG, PNG, WebP or GIF; videos can be MP4, MOV or WebM.
• list_media returns ids, sizes and dimensions of what is already uploaded, never a download link. (The REST GET /api/v1/media does include one.)
• A post takes up to 20 media ids, attached in the order you give them.
• Larger files: upload them at ihateposting.com and find them with list_media, or use the REST upload routes, which take images up to 25 MB and videos up to 512 MB in one upload, or up to 2,000 MB in parts. See docs/api.md.
• Your library has a storage quota: 500 MB on the 90-day free trial. The paid plans, once they go on sale, have 5 GB (Creator), 25 GB (Growth) and 100 GB (Pro).
• Alt text is published on Bluesky, Instagram, Threads, Mastodon, Pinterest, Tumblr and Slack. LinkedIn uses it as a video's title.
• A custom video cover is an image from your library, set with coverAssetId in that network's options. YouTube, Instagram, Facebook, LinkedIn, Pinterest, Telegram, Mastodon and Tumblr use it. TikTok does not take an uploaded cover.
• TikTok photo posts take JPEG or WebP, not PNG.

Analytics

get_analytics takes a range of 7d, 15d, 30d (the default), 90d, 12m or all. For each connected channel it returns that network's own metrics, follower numbers, views, engagements and engagement rate, posting times by weekday and time of day, a daily trend, and the top five posts. Analytics is the same on every plan, including long ranges. Over the REST API, GET /api/v1/analytics?range=30d&format=xlsx returns a spreadsheet of the posts you published through iHatePosting in that range.

Per-post numbers come from Bluesky, Facebook, Instagram, Threads, LinkedIn Pages, Pinterest, TikTok (public posts only), YouTube, Mastodon, Tumblr, Slack, and Discord channels connected through the bot. Telegram reports no per-post numbers, only a member count, and a LinkedIn personal profile is not measured. X is measured only when the service has X read access switched on. The answer's unmeasuredPlatforms field names each connected network it does not measure at all, so a missing number is not mistaken for no reach.

Other ways to connect

Local MCP server (npm). For clients that start a local process, the same 16 tools run from ihateposting-mcp, which needs Node.js 20 or later (since 0.10.0):

{  "mcpServers": {    "ihateposting": {      "command": "npx",      "args": ["-y", "ihateposting-mcp"],      "env": { "IHATEPOSTING_API_KEY": "pk_live_…" }    }  }}

IHATEPOSTING_API_URL changes the server it talks to (default https://ihateposting.com).

Command line. ihateposting (npm i -g ihateposting, or npx ihateposting <command> without installing) drafts, checks, schedules and publishes, uploads files off disk, and lists posts. Its commands are login, logout, whoami, accounts, platforms, posts, upload, post and skill; ihateposting help lists the commands and the main flags.

ihateposting login <key>                                   # saved in ~/.ihateposting/config.jsonihateposting whoami                                        # which account the key belongs toihateposting accounts --json                               # connected accounts, with idsihateposting upload ./launch.mp4 --alt "Product demo"      # prints a media idihateposting post "text" --to bluesky,linkedin --check     # validates this draft, creates nothingihateposting post "text" --to instagram --media "$ID"      # draft with the uploaded fileihateposting post "text" --to x --at "2026-10-01 9:00 AM" --check   # check the scheduled version firstihateposting post "text" --to x --at "2026-10-01 9:00 AM"  # scheduledihateposting posts --status scheduled --from 2026-12-01 --to 2026-12-31   # prints the total tooihateposting skill                                         # writes .claude/skills/ihateposting/SKILL.md

--check validates the command exactly as written, so give it the same --now or --at you will post with. --now publishes at once. The command line cannot set per-network options yet, so a Pinterest board or YouTube's made-for-kids answer needs the MCP tools or the REST API. Every command exits 0 on success and 1 on failure. ihateposting skill --print prints the CLI's own skill instead of writing it.

To give the CLI's skill to any agent — Claude Code, Cursor, Codex, GitHub Copilot and the rest — install it from this repository with the open skills installer, which asks which agents to set it up for:

npx skills add iHatePostingOrg/agent-skill --skill ihateposting-cli

It is cli/ihateposting-cli/SKILL.md: the same guidance ihateposting skill writes, for an agent that posts with the command line. It is not part of any plugin above. The settings the command line and the npm server read are listed in their own READMEs on npm (ihateposting, ihateposting-mcp).

A URL with the key in it. Some connector dialogs accept only a URL, with no field for a header. For those, the server also reads the key from the address: https://ihateposting.com/mcp?key=pk_live_…. A key in a URL can end up in browser history, proxy logs and Referer headers, so use the header whenever the client allows one, and make a new key if a URL containing it is ever shared.

MCP Registry. The server is listed in the official MCP Registry as com.ihateposting/mcp, with both addresses: https://ihateposting.com/mcp/oauth, where a client signs in, and https://ihateposting.com/api/mcp (the same server as /mcp), which takes a key as Authorization: Bearer. The entry also names the local npm server, ihateposting-mcp.

REST API and webhooks

The MCP tools are a thin layer over a public REST API, which takes the same key as Authorization: Bearer pk_live_…:

Request and response shapes are in docs/api.md.

Webhooks tell another system when a post is done. Add them at ihateposting.com under Settings → Webhooks (up to 10). Each receives a signed POST for post.published, post.partial or post.failed:

• The X-iHatePosting-Signature header is sha256= followed by the HMAC-SHA256 of the raw body, keyed with that webhook's secret (whsec_…).
• The payload's id stays the same when a delivery is retried, so use it to drop duplicates.
• A webhook can be limited to chosen accounts. It then fires for posts that touched one of them and carries only those accounts' results.
• Send test delivers a sample payload of the same shape, marked with an X-iHatePosting-Test: true header.

The setup for Zapier, Make and n8n is in this guide.

Limits

• Requests per minute, per account: 30 creates, 60 updates, 30 reschedules, 15 retries, 60 validations, 60 analytics reads, 120 media listings, 60 Pinterest board lookups and 30 uploads. list_posts, list_accounts, whoami and get_platform_rules share 60 a minute. Going over returns HTTP 429; wait a minute and try again.
• One post per create_post call, with up to 20 media ids.
• list_posts is paged. 50 by default, 200 maximum. The reply carries total (matching your filter) and returned (this page), plus hasMore and nextCursor. Take counts from total, never from the page — and filter with status and from/to rather than counting rows.
• The 90-day free trial. Paid plans are not on sale yet; plans and what each includes are on the pricing page linked below.
• X is available on every plan within a monthly number of posts that depends on the plan, and a daily number. Links in X posts need a paid Pro plan; during the free trial, post to X without the link or give X its own text.

Plans and prices: ihateposting.com/pricing.

Troubleshooting

• The tools are missing. The agent is not signed in. On claude.ai and in Cowork, connect iHatePosting from the plugin's Connectors tab; in Claude Code, run /mcp and choose Authenticate; in Cursor, open Customize, then MCPs, turn iHatePosting on and press Connect; in Gemini CLI, run /mcp auth ihateposting; in Grok Build, open /mcps and press i; in Kimi Code, approve its authenticate step or run /mcp-config login with the name /mcp shows; in Qwen Code, open /mcp, select iHatePosting and choose Authenticate. None of these use a key.
• Coming from 0.4.x on Cursor or Grok Build? Those plugins used to send an API key. They now sign in instead, once, as above. The plugin no longer reads the key you set for 0.4.x, so you can remove it from Cursor's plugin settings or your shell profile. If nothing else uses that key, regenerate it under Settings → Developers so the old one stops working.
• "No API key" from the command line or the npm server. It was given no key. Run ihateposting login <key>.
• "iHatePosting API 401" on a key-based setup. The key is wrong, has been replaced by a newer one, or the client sent an unfilled placeholder instead of the key. Check the client's setting before assuming the key was revoked.
• An account shows needs_reauth. It has to be reconnected on the Accounts page at ihateposting.com; an agent cannot do that. Its pending sends are held, and go out once the same account is reconnected. A send still held 14 days after it was due fails instead; reconnect, then retry it.
• HTTP 429. A per-minute limit was reached (see Limits).
• Posted to fewer networks than asked. The create_post answer lists every requested name or id that matched no connected, active account under unresolved.
• Some networks published and others failed. The post's status is partial. get_post shows each network's error, and retry_post resends only the failed ones. A send can also be skipped when the publisher's last check refused it; fix the post rather than retrying.
• Refused when creating or updating. The message is the network's own rule, for example a caption over the limit or missing media. Fix it, or give that network an override, and send again. A draft skips these checks.
• Refused for an option ("did you mean ytPrivacy?"). The key or value is not one that network takes. Use the one the message names; get_platform_rules lists every option. This check runs on drafts too.
• YouTube is refused. It needs one video, a title, and the made-for-kids declaration: set ytMadeForKids to true or false in the youtube options. Privacy is public unless you set ytPrivacy.
• Pinterest is refused. A pin needs a board. Get the id from list_pinterest_boards and set it as boardId in the pinterest options. An empty list means the account has no boards yet; make one on Pinterest first.
• TikTok went to the wrong audience, or not at all. Without tiktokPrivacy, a post sent through the agent or the API goes to the widest audience TikTok offers that account, which is public for a public account. Set it to public, followers, friends or private; any other word is refused. With tiktokSendAsDraft, the video goes to the creator's TikTok inbox and is reported as done, but it is only public once they post it in the TikTok app.

Network endpoints and credentials

• Every agent in this package calls one address, https://ihateposting.com/mcp/oauth: MCP over Streamable HTTP, with OAuth sign-in. It sends no key. Each agent discovers the sign-in from the server's own metadata, uses PKCE and registers itself, then keeps its own token (Gemini CLI in ~/.gemini/mcp-oauth-tokens.json, Grok Build in ~/.grok/mcp_credentials.json, Kimi Code under ~/.kimi-code/credentials/mcp/, Qwen Code in ~/.qwen/mcp-oauth-tokens.json). Revoke a sign-in at ihateposting.com under Settings → Developers.
• https://ihateposting.com/mcp is the same server for key-based setups (the local npm server, agents that cannot sign in), which send Authorization: Bearer <key>. Nothing in this package uses it.
• There are no hooks and no install scripts, and nothing runs on your machine.
• What the server receives: the text, media and options of the posts you ask for, and the requests to list, check, change or report on them. It sends them on to the social networks you chose, through each network's official API. The privacy policy is at ihateposting.com/privacy.

What is in this repository

Each agent has its own MCP file, so that each manifest names exactly one server; all of them point at the same sign-in address. There is deliberately no .mcp.json at the root, so no tool picks up a server this package did not mean to give it.

node scripts/check.mjs checks the manifests, the key handling, the skill and the examples before a release. It also checks that this README names every tool and network, and that every example is a draft.

Feedback and issues

Report a problem or ask for something in this repository's issues, or write to us from ihateposting.com/contact.

Links

• Website: ihateposting.com
• Set-up pages for every agent: ihateposting.com/ai-agents
• Setup guide: Post to social media from an AI agent
• REST API: docs/api.md
• Pricing: ihateposting.com/pricing
• Privacy policy: ihateposting.com/privacy
• Terms: ihateposting.com/terms

Changelog

See CHANGELOG.md.

License

MIT. See LICENSE.