<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Dominic Frei</title>
    <description>The latest articles on DEV Community by Dominic Frei (@dominic_frei_73028ac46c47).</description>
    <link>https://dev.to/dominic_frei_73028ac46c47</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4142432%2Fccfd877f-a191-4f6d-9bd9-728c2a5adcc3.png</url>
      <title>DEV Community: Dominic Frei</title>
      <link>https://dev.to/dominic_frei_73028ac46c47</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/dominic_frei_73028ac46c47"/>
    <language>en</language>
    <item>
      <title>How to connect Claude to social media with MCP</title>
      <dc:creator>Dominic Frei</dc:creator>
      <pubDate>Fri, 20 Nov 2026 13:16:58 +0000</pubDate>
      <link>https://dev.to/dominic_frei_73028ac46c47/how-to-connect-claude-to-social-media-with-mcp-3eg6</link>
      <guid>https://dev.to/dominic_frei_73028ac46c47/how-to-connect-claude-to-social-media-with-mcp-3eg6</guid>
      <description>&lt;p&gt;To connect Claude to social media, give it an MCP server that holds your platform credentials and exposes posting as tools. Claude never signs in to Instagram or X itself; it calls tools, and the server publishes. With pendpost the setup is one stdio entry, &lt;code&gt;npx -y pendpost --stdio&lt;/code&gt;, added to Claude Desktop, Claude Code or Cursor. Start with &lt;code&gt;PENDPOST_MODE=mock&lt;/code&gt; and a temp-folder &lt;code&gt;PENDPOST_ROOT&lt;/code&gt;, have Claude draft a post, approve it yourself, and watch the mock driver publish it with zero real API calls. Real accounts come after that, one platform at a time.&lt;/p&gt;

&lt;h2&gt;
  
  
  What does Claude need to post to social media?
&lt;/h2&gt;

&lt;p&gt;Claude needs an MCP server that holds the platform tokens and decides when a post may go out. Claude Desktop, Claude Code and Cursor are MCP clients. They launch local servers, list their tools, and let the model call them. The MCP specification's &lt;a href="https://modelcontextprotocol.io/specification/2025-06-18/server/tools" rel="noopener noreferrer"&gt;tools page&lt;/a&gt; says there should always be a human in the loop who can deny tool invocations, and Claude Desktop asks for your approval before it runs one.&lt;/p&gt;

&lt;p&gt;pendpost is one such server. It is MIT-licensed, binds 127.0.0.1, and keeps tokens in your own &lt;code&gt;.env&lt;/code&gt;; every post passes its approval gate. The &lt;a href="https://pendpost.com/mcp" rel="noopener noreferrer"&gt;MCP server page&lt;/a&gt; lists what the tools cover. Everything below works the same for any stdio MCP server, only the command changes.&lt;/p&gt;

&lt;h2&gt;
  
  
  Which config works for Claude Desktop, Claude Code and Cursor?
&lt;/h2&gt;

&lt;p&gt;The config that works is the plain stdio launch plus three environment variables. Here it is for each client, as of November 2026.&lt;/p&gt;

&lt;h3&gt;
  
  
  Claude Desktop
&lt;/h3&gt;

&lt;p&gt;Open Settings from the Claude menu in your system menu bar, go to Developer, and click Edit Config. The file lives at &lt;code&gt;~/Library/Application Support/Claude/claude_desktop_config.json&lt;/code&gt; on macOS and &lt;code&gt;%APPDATA%\Claude\claude_desktop_config.json&lt;/code&gt; on Windows, per the &lt;a href="https://modelcontextprotocol.io/docs/develop/connect-local-servers" rel="noopener noreferrer"&gt;MCP docs&lt;/a&gt;.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight json"&gt;&lt;code&gt;&lt;span class="p"&gt;{&lt;/span&gt;
  &lt;span class="nl"&gt;"mcpServers"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;
    &lt;span class="nl"&gt;"pendpost"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;
      &lt;span class="nl"&gt;"command"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"npx"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
      &lt;span class="nl"&gt;"args"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="s2"&gt;"-y"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="s2"&gt;"pendpost"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="s2"&gt;"--stdio"&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
      &lt;span class="nl"&gt;"env"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;
        &lt;span class="nl"&gt;"PENDPOST_MODE"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"mock"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
        &lt;span class="nl"&gt;"PENDPOST_ROOT"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"/tmp/pendpost-demo"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
        &lt;span class="nl"&gt;"PENDPOST_PORT"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"8197"&lt;/span&gt;
      &lt;span class="p"&gt;}&lt;/span&gt;
    &lt;span class="p"&gt;}&lt;/span&gt;
  &lt;span class="p"&gt;}&lt;/span&gt;
&lt;span class="p"&gt;}&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Quit Claude Desktop completely and reopen it. A config change only loads on restart.&lt;/p&gt;

&lt;h3&gt;
  
  
  Claude Code
&lt;/h3&gt;

&lt;p&gt;One command, run from your project folder. The &lt;code&gt;--&lt;/code&gt; separator matters: everything after it goes to the server unchanged (&lt;a href="https://code.claude.com/docs/en/mcp" rel="noopener noreferrer"&gt;Claude Code MCP docs&lt;/a&gt;).&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;claude mcp add --env &lt;span class="nv"&gt;PENDPOST_MODE&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;mock --env &lt;span class="nv"&gt;PENDPOST_ROOT&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;/tmp/pendpost-demo --env &lt;span class="nv"&gt;PENDPOST_PORT&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;8197 --transport stdio pendpost -- npx -y pendpost --stdio
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Keep &lt;code&gt;--transport stdio&lt;/code&gt; between the last &lt;code&gt;--env&lt;/code&gt; and the name, or the CLI reads &lt;code&gt;pendpost&lt;/code&gt; as another variable. The default scope is local to the project. Add &lt;code&gt;--scope user&lt;/code&gt; to load it everywhere, or &lt;code&gt;--scope project&lt;/code&gt; to write a shared &lt;code&gt;.mcp.json&lt;/code&gt;. Run &lt;code&gt;/mcp&lt;/code&gt; inside Claude Code to see the connection.&lt;/p&gt;

&lt;h3&gt;
  
  
  Cursor
&lt;/h3&gt;

&lt;p&gt;Put the same &lt;code&gt;mcpServers&lt;/code&gt; block in &lt;code&gt;.cursor/mcp.json&lt;/code&gt; in your project or &lt;code&gt;~/.cursor/mcp.json&lt;/code&gt; for all projects. Cursor's &lt;a href="https://cursor.com/docs/context/mcp" rel="noopener noreferrer"&gt;MCP docs&lt;/a&gt; list &lt;code&gt;"type": "stdio"&lt;/code&gt; as a field, so add it next to &lt;code&gt;command&lt;/code&gt;.&lt;/p&gt;

&lt;h3&gt;
  
  
  Any other stdio client
&lt;/h3&gt;

&lt;p&gt;Any client that can launch a stdio server takes the same command, args and env. If you prefer one click, the &lt;code&gt;.mcpb&lt;/code&gt; bundle on the &lt;a href="https://pendpost.com/download" rel="noopener noreferrer"&gt;download page&lt;/a&gt; installs into Claude Desktop without editing JSON, but it starts in live mode, so use the config above for a first mock run.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why set PENDPOST_ROOT and PENDPOST_PORT?
&lt;/h2&gt;

&lt;p&gt;You set them because of two behaviours that otherwise look like a silent failure. Both come from the engine code, and both made my own test runs exit before the first tool call.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Variable&lt;/th&gt;
&lt;th&gt;Value&lt;/th&gt;
&lt;th&gt;What happens without it&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;PENDPOST_MODE&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;&lt;code&gt;mock&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;pendpost runs live. Unconnected lanes wait for a credential and nothing fakes a publish, but nothing publishes either.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;PENDPOST_ROOT&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;a folder under &lt;code&gt;/tmp&lt;/code&gt;
&lt;/td&gt;
&lt;td&gt;In mock mode, pendpost refuses to boot against a data folder outside the temp directory, so fake post ids can never land in real plans. On 2.8.0 and earlier that included the default folder, so it logged "mock mode refused" and exited with code 1. Newer versions fall back to a private demo folder in your temp directory; setting it yourself keeps the demo somewhere you can find it.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;PENDPOST_PORT&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;any free port, such as &lt;code&gt;8197&lt;/code&gt;
&lt;/td&gt;
&lt;td&gt;The stdio process also starts the dashboard, on 8090 by default. If another pendpost holds 8090, the new one logs "port 8090 already in use" and exits with code 0, so the client just shows a disconnected server.&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;On macOS and Linux, &lt;code&gt;/tmp/pendpost-demo&lt;/code&gt; passes the temp check; the folder does not have to exist first. On Windows, use a folder inside the directory that &lt;code&gt;node -p "require('os').tmpdir()"&lt;/code&gt; prints. Node.js 22 or newer is required.&lt;/p&gt;

&lt;p&gt;If you already run pendpost as a background app on 8090 and want Claude to drive that same instance, skip stdio. In Claude Code, &lt;code&gt;claude mcp add --transport http pendpost http://127.0.0.1:8090/mcp&lt;/code&gt; connects to it directly. The &lt;a href="https://docs.pendpost.com/mcp-server" rel="noopener noreferrer"&gt;MCP clients docs page&lt;/a&gt; covers the HTTP route for Claude Desktop too.&lt;/p&gt;

&lt;h2&gt;
  
  
  What should I ask Claude first?
&lt;/h2&gt;

&lt;p&gt;Start with prompts that each map to one tool call, so you can see what happened. I ran this exact sequence over stdio against pendpost 2.8.0 in mock mode; the responses below are what came back.&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;"Call pendpost_health and tell me what is set up." It returns &lt;code&gt;ready: false&lt;/code&gt; with one "not yet proven" line per platform. That is correct for a fresh mock run: no lane has a real credential.&lt;/li&gt;
&lt;li&gt;"Create a campaign called mcp-test." A fresh temp root starts empty, so this is the first write. Claude calls &lt;code&gt;campaign_create&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;"Draft a Mastodon text post in mcp-test, due now, that says we are testing a draft from my agent." Claude calls &lt;code&gt;plan_create_post&lt;/code&gt;. The post comes back with &lt;code&gt;approval: "draft"&lt;/code&gt; and &lt;code&gt;createdBy&lt;/code&gt; set to the agent's actor name, for example &lt;code&gt;agent:claude&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;"Publish the due posts now." Nothing publishes. &lt;code&gt;publish_due_run&lt;/code&gt; returns an empty list, because a draft is not publishable.&lt;/li&gt;
&lt;li&gt;"Approve it." If Claude approves under its own actor name, pendpost answers "agent:claude created this post and cannot approve it (no self-approval)". Approve it yourself in the dashboard at &lt;code&gt;http://127.0.0.1:8197&lt;/code&gt;, or tell Claude in plain words that you, the owner, approve it.&lt;/li&gt;
&lt;li&gt;"Publish the due posts now." This time the run reports &lt;code&gt;ok: true&lt;/code&gt; for the Mastodon lane, and the post's status turns to &lt;code&gt;posted&lt;/code&gt;. The mock driver invented the post id and wrote the action to a &lt;code&gt;.mock-ledger.json&lt;/code&gt; file inside your temp folder.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;The &lt;a href="https://docs.pendpost.com/mock-mode" rel="noopener noreferrer"&gt;mock mode docs&lt;/a&gt; describe what the mock driver fakes and what it skips (rate limits and native scheduling only engage on a live lane).&lt;/p&gt;

&lt;h2&gt;
  
  
  What stops Claude from posting something I did not approve?
&lt;/h2&gt;

&lt;p&gt;The approval gate does: every post starts as a draft, only the approve and reject tools (or an auto-approve policy that only the owner can switch on) change that, and the actor that drafted a post cannot approve it. A human, as owner, can approve their own drafts. If the copy changes after approval, the post is flagged as edited and waits for a fresh approval. In my test run, the publish sweep skipped it.&lt;/p&gt;

&lt;p&gt;One honest limit. The actor is a name the caller passes in, so the rule stops an agent from blessing its own draft under its own name; it does not check who is behind the name "owner". The second checkpoint is your MCP client, which asks before running a tool. Read the &lt;code&gt;approve_post&lt;/code&gt; prompt before you allow it, and do not set it to always allow. For the wider pattern, see &lt;a href="https://pendpost.com/blog/let-an-ai-agent-post-to-social-media-safely" rel="noopener noreferrer"&gt;how to let an AI agent post to social media safely&lt;/a&gt;, and the &lt;a href="https://pendpost.com/security" rel="noopener noreferrer"&gt;security page&lt;/a&gt; for what stays on your machine.&lt;/p&gt;

&lt;h2&gt;
  
  
  How do I move from mock mode to real accounts?
&lt;/h2&gt;

&lt;p&gt;Remove &lt;code&gt;PENDPOST_MODE&lt;/code&gt;, point &lt;code&gt;PENDPOST_ROOT&lt;/code&gt; at a permanent folder, and connect one platform at a time. Ask Claude to call &lt;code&gt;pendpost_health&lt;/code&gt; again: its &lt;code&gt;setup&lt;/code&gt; field lists, per platform, what is missing and the exact command to run. Claude can write non-secret identifiers such as a page id, but you run the sign-in step yourself, in your terminal or the dashboard Setup page, so the token goes into your &lt;code&gt;.env&lt;/code&gt; and never through the chat. The &lt;a href="https://docs.pendpost.com/quickstart" rel="noopener noreferrer"&gt;quickstart&lt;/a&gt; links the credential guide for each platform.&lt;/p&gt;

&lt;p&gt;Approve your first live post by hand. It also pays to keep the mock entry in your config under a second name, such as &lt;code&gt;pendpost-mock&lt;/code&gt;, with its own port and temp folder. A new prompt can then run against fake post ids first, which cost nothing when they go wrong.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Originally published on &lt;a href="https://pendpost.com/blog/connect-claude-to-social-media-mcp" rel="noopener noreferrer"&gt;pendpost.com&lt;/a&gt;. pendpost is a free, open-source (MIT), local-first social media planner where an AI agent drafts posts behind a human approval gate.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>automation</category>
      <category>claude</category>
      <category>mcp</category>
      <category>socialmedia</category>
    </item>
    <item>
      <title>Social media API posting limits in 2026: every platform</title>
      <dc:creator>Dominic Frei</dc:creator>
      <pubDate>Fri, 20 Nov 2026 12:56:11 +0000</pubDate>
      <link>https://dev.to/dominic_frei_73028ac46c47/social-media-api-posting-limits-in-2026-every-platform-1okf</link>
      <guid>https://dev.to/dominic_frei_73028ac46c47/social-media-api-posting-limits-in-2026-every-platform-1okf</guid>
      <description>&lt;p&gt;As of November 2026, the hard API posting caps are: Instagram 100 posts per rolling 24 hours, Facebook Page Reels 30, Threads 250 posts plus 1,000 replies, YouTube 100 uploads per day from a separate quota bucket, X 100 posts per 15 minutes per user, and Bluesky about 1,666 record writes per hour. Mastodon's default server code allows 300 statuses per 3 hours. TikTok gives a creator roughly 15 Direct Post uploads a day. LinkedIn, Facebook Page feed posts and Pinterest publish no fixed number at all. The full table is below, with text and media limits, followed by the places where our production code disagrees with the docs.&lt;/p&gt;

&lt;h2&gt;
  
  
  What are the posting limits for each social media API?
&lt;/h2&gt;

&lt;p&gt;The table below has the documented number for each platform, as of November 2026. I read every cell off the platform's own docs in the last week of October 2026. Where a platform publishes no number, the cell says "not published".&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Platform&lt;/th&gt;
&lt;th&gt;Posting cap&lt;/th&gt;
&lt;th&gt;Text limit&lt;/th&gt;
&lt;th&gt;Media and carousel limits&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Instagram (Graph API)&lt;/td&gt;
&lt;td&gt;100 API-published posts per 24 h moving window; carousel counts as 1&lt;/td&gt;
&lt;td&gt;2,200 characters, 30 hashtags, 20 @ tags&lt;/td&gt;
&lt;td&gt;Carousel up to 20 items; Reels 3 s to 15 min, 300 MB; feed images JPEG only, 8 MB; Stories video up to 60 s, 100 MB&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Facebook Pages&lt;/td&gt;
&lt;td&gt;Reels: 30 per 24 h. Feed and photo posts: not published&lt;/td&gt;
&lt;td&gt;Not published&lt;/td&gt;
&lt;td&gt;Reels 3 to 90 s, 9:16; photos 4 MB, JPEG, PNG, GIF, BMP or TIFF&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Threads&lt;/td&gt;
&lt;td&gt;250 posts and 1,000 replies per 24 h&lt;/td&gt;
&lt;td&gt;500 characters, at most 5 links&lt;/td&gt;
&lt;td&gt;Carousel 2 to 20 items; video up to 5 min, 1 GB&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;LinkedIn&lt;/td&gt;
&lt;td&gt;Not published; per-app and per-member daily limits shown in the Developer Portal&lt;/td&gt;
&lt;td&gt;3,000 characters (Help Center)&lt;/td&gt;
&lt;td&gt;MultiImage 2 to 20 images; video 3 s to 30 min, MP4, 75 KB to 500 MB&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;YouTube (Data API v3)&lt;/td&gt;
&lt;td&gt;Uploads: 100 per day at 1 unit each, own bucket; all other calls: 10,000 units per day&lt;/td&gt;
&lt;td&gt;Title 100 characters, description 5,000 bytes, tags 500 characters&lt;/td&gt;
&lt;td&gt;File up to 256 GB&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;X (API v2)&lt;/td&gt;
&lt;td&gt;100 posts per 15 min per user, 10,000 per 24 h per app; billed per post&lt;/td&gt;
&lt;td&gt;280 weighted characters; 25,000 with Premium&lt;/td&gt;
&lt;td&gt;4 photos, or 1 GIF, or 1 video; image 5 MB, GIF 15 MB; video 20 min and 8 GB (Premium: 125 min, 16 GB)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Mastodon (default server)&lt;/td&gt;
&lt;td&gt;300 statuses per 3 h per account; 300 API calls per 5 min; 30 media uploads per 30 min&lt;/td&gt;
&lt;td&gt;500 characters, each URL counts as 23&lt;/td&gt;
&lt;td&gt;4 attachments, at most 1 video; 16 MB image, 99 MB video in the current source&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Bluesky&lt;/td&gt;
&lt;td&gt;5,000 points per hour and 35,000 per day; a create costs 3 points&lt;/td&gt;
&lt;td&gt;300 graphemes (3,000 bytes)&lt;/td&gt;
&lt;td&gt;4 images at 2 MB each; video 300 MB&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Telegram (Bot API)&lt;/td&gt;
&lt;td&gt;About 30 messages per second overall, 1 per second per chat, 20 per minute per group&lt;/td&gt;
&lt;td&gt;4,096 characters per message, 1,024 per media caption&lt;/td&gt;
&lt;td&gt;Album of 2 to 10 items; uploads 10 MB for photos, 50 MB for other files&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Discord (webhooks)&lt;/td&gt;
&lt;td&gt;Not published per route; read the rate limit headers. 50 requests per second per bot globally&lt;/td&gt;
&lt;td&gt;2,000 characters&lt;/td&gt;
&lt;td&gt;Up to 10 embeds; files up to 20 MiB by default&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Pinterest (API v5)&lt;/td&gt;
&lt;td&gt;Not published in the API description&lt;/td&gt;
&lt;td&gt;Title 100, description 800, alt text 500 characters&lt;/td&gt;
&lt;td&gt;Multi-image pin 2 to 5 images&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;TikTok (Content Posting API)&lt;/td&gt;
&lt;td&gt;About 15 posts per day per creator; 6 init requests per minute per token&lt;/td&gt;
&lt;td&gt;2,200 UTF-16 runes&lt;/td&gt;
&lt;td&gt;Video up to 10 min and 4 GB via upload; photos 20 MB, JPEG or WebP&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Two cells need a footnote. Meta's &lt;a href="https://developers.facebook.com/docs/instagram-platform/content-publishing" rel="noopener noreferrer"&gt;content publishing guide&lt;/a&gt; says 100 posts in its rate limit section and 50 in its carousel section. And the YouTube upload number changed twice in a year: the &lt;a href="https://developers.google.com/youtube/v3/revision_history" rel="noopener noreferrer"&gt;revision history&lt;/a&gt; records an upload dropping from about 1,600 units to about 100 in December 2025, then a move to its own bucket on June 1, 2026. If your quota math is from 2024, redo it.&lt;/p&gt;

&lt;h2&gt;
  
  
  Which limits will you hit first?
&lt;/h2&gt;

&lt;p&gt;The first limit most people hit is not a daily cap. It is a format rule: a 281-character post on X, a 4.5 MB photo on a Facebook Page, an eleventh carousel slide. Daily caps only bite when you automate at volume, and at that point quiet failures cost more than the cap does.&lt;/p&gt;

&lt;p&gt;A few rules sit outside the table and catch people anyway.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;X does not count the way &lt;code&gt;string.length&lt;/code&gt; does. Per the &lt;a href="https://docs.x.com/fundamentals/counting-characters" rel="noopener noreferrer"&gt;counting characters guide&lt;/a&gt;, every URL counts as 23 characters and every emoji as 2. A tweet with a long tracking link can be shorter than it looks, and one full of emoji can be longer.&lt;/li&gt;
&lt;li&gt;X bills per call. On pay-per-use &lt;a href="https://docs.x.com/x-api/getting-started/pricing" rel="noopener noreferrer"&gt;pricing&lt;/a&gt;, a post costs $0.015 and a post containing a URL costs $0.20. When the credits run out, every request fails until you top up (in our logs, as HTTP 402).&lt;/li&gt;
&lt;li&gt;TikTok hides posts from unaudited apps. The &lt;a href="https://developers.tiktok.com/doc/content-sharing-guidelines" rel="noopener noreferrer"&gt;content sharing guidelines&lt;/a&gt; restrict an unaudited app to private (&lt;code&gt;SELF_ONLY&lt;/code&gt;) posts and to 5 posting users per 24 hours.&lt;/li&gt;
&lt;li&gt;Mastodon limits belong to each server. 500 characters and 4 attachments are the defaults in Mastodon's source, but an admin can change them. Read &lt;code&gt;configuration.statuses&lt;/code&gt; from the &lt;a href="https://docs.joinmastodon.org/entities/Instance/" rel="noopener noreferrer"&gt;instance entity&lt;/a&gt; before you trust a number.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  What do the docs get wrong in practice?
&lt;/h2&gt;

&lt;p&gt;Some limits only show up when you publish for real. These come from running pendpost's own lanes every day.&lt;/p&gt;

&lt;p&gt;Instagram's resumable upload gives up at about 35 seconds. The docs allow Reels up to 15 minutes, and that is true, just not for every upload path. In October 2026 we timed the resumable upload (&lt;code&gt;rupload&lt;/code&gt;) on a live account, with containers we never published. Uploads that passed returned within 26 seconds. Longer reels failed at 35 to 37 seconds with &lt;code&gt;HTTP 400 ProcessingFailedError&lt;/code&gt;, whatever the file size or encoding, and the container then stayed &lt;code&gt;IN_PROGRESS&lt;/code&gt; for good. The same 102 MB, 70-second file sent through &lt;code&gt;video_url&lt;/code&gt; (a public HTTPS link Meta fetches itself) finished in 64 seconds. My reading: the resumable path processes the video inside the request and Meta cuts the request off. A long reel needs &lt;code&gt;video_url&lt;/code&gt;. For weeks it looked like random Meta flakiness.&lt;/p&gt;

&lt;p&gt;The "25 posts per day" Instagram figure is out of date. It still circulates, but the current guide says 100. We dropped our own conservative default (2 posts, 6 hours apart) for that reason.&lt;/p&gt;

&lt;p&gt;X refuses over-length posts from the API too. The long posts you see on x.com are a Premium feature, and the API gives a non-Premium account no way around 280. A scheduler that skips an over-length post without recording a failure leaves it overdue forever with no error anywhere. We shipped exactly that bug once.&lt;/p&gt;

&lt;p&gt;YouTube quota errors look like permission errors. A quota 403 (&lt;code&gt;quotaExceeded&lt;/code&gt;, &lt;code&gt;rateLimitExceeded&lt;/code&gt;) has the same status code as a missing OAuth scope. Telling the user to reconnect fixes nothing; waiting for the midnight Pacific reset does.&lt;/p&gt;

&lt;h2&gt;
  
  
  How does pendpost encode these limits?
&lt;/h2&gt;

&lt;p&gt;pendpost keeps each limit in one place that the pre-approval check and the publish engine both read, so a post that fails the limit is flagged before you approve it, not after it is due. The numbers live in &lt;code&gt;lib/caption.mjs&lt;/code&gt;, &lt;code&gt;lib/carousel.mjs&lt;/code&gt; and the scheduler in the &lt;a href="https://github.com/pendpost/pendpost" rel="noopener noreferrer"&gt;open-source repo&lt;/a&gt;.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Lane&lt;/th&gt;
&lt;th&gt;What pendpost enforces&lt;/th&gt;
&lt;th&gt;Where it differs from the docs&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Instagram and Facebook&lt;/td&gt;
&lt;td&gt;One shared cap of 100 posts per rolling 24 h, with a warning at 80 percent; a capped post waits, it is never dropped&lt;/td&gt;
&lt;td&gt;The cap is shared across both, and Facebook Reels have their own lower cap of 30&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Instagram carousel&lt;/td&gt;
&lt;td&gt;10 items in pendpost 2.8.0; 20 from the next release&lt;/td&gt;
&lt;td&gt;Meta documents 20; the fix is on the main branch and ships in the next release&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Facebook photos&lt;/td&gt;
&lt;td&gt;4 MB and JPEG, PNG, GIF, BMP or TIFF, checked before any upload; 10 photos per album&lt;/td&gt;
&lt;td&gt;Meta documents no album count, so 10 is our own cautious cap&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;X&lt;/td&gt;
&lt;td&gt;280 is a hard error at approval time; over-length text can be split into a thread&lt;/td&gt;
&lt;td&gt;Matches X for non-Premium accounts; Premium long posts are not supported. In 2.8.0 approval counts URLs as 23 but publish counts raw length; the next release uses X's own count (URLs 23, emoji 2) in both places&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;YouTube&lt;/td&gt;
&lt;td&gt;Description capped at 5,000 characters in 2.8.0&lt;/td&gt;
&lt;td&gt;YouTube counts 5,000 bytes, so accented text can pass our check and be refused. The next release counts bytes, checks the 100-character title, &amp;lt; and &amp;gt;, and tags the way YouTube does&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Mastodon&lt;/td&gt;
&lt;td&gt;500 characters including the content warning; 4 attachments, one video&lt;/td&gt;
&lt;td&gt;Matches the default server; a custom instance limit is not read&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Telegram, Discord, Pinterest, TikTok&lt;/td&gt;
&lt;td&gt;1,024 caption or 4,096 text; 2,000; title 100 and description 800; 2,200&lt;/td&gt;
&lt;td&gt;Match the docs&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Writing this table turned up three gaps in pendpost 2.8.0: the carousel cap, the YouTube byte count, and X text counted two different ways. All three are fixed on the main branch and ship in the next release; the shared Meta cap is still open. I would rather list them here than have you find them in production. Beyond the limits, a Meta action block (error 368) or an X credits error halts that lane until you clear it, instead of retrying into a suspension. The &lt;a href="https://docs.pendpost.com/approval-and-circuit-breakers" rel="noopener noreferrer"&gt;approval and circuit breaker docs&lt;/a&gt; cover how that works, and the &lt;a href="https://docs.pendpost.com/credentials/meta" rel="noopener noreferrer"&gt;Meta credentials page&lt;/a&gt; covers the Instagram and Facebook setup.&lt;/p&gt;

&lt;h2&gt;
  
  
  How should you plan around platform limits?
&lt;/h2&gt;

&lt;p&gt;Check limits when a post is created, not when it is due, and read live quotas where the platform offers them. In practice that comes down to this:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Validate at draft time. Count characters the way each platform counts them, and check media size, format and carousel length before anyone approves the post.&lt;/li&gt;
&lt;li&gt;Read live quotas. Instagram exposes &lt;code&gt;content_publishing_limit&lt;/code&gt; and Threads exposes &lt;code&gt;threads_publishing_limit&lt;/code&gt;; LinkedIn shows usage in the Developer Portal. Trust those over any table, this one included.&lt;/li&gt;
&lt;li&gt;When a cap is reached, keep the post due and try again later, and log one row for it instead of one per retry.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;All of this matters more once an agent does the scheduling. An agent will happily queue 40 posts in a minute and never wonder about Meta's carousel section. The guide on &lt;a href="https://pendpost.com/blog/let-an-ai-agent-post-to-social-media-safely" rel="noopener noreferrer"&gt;letting an AI agent post safely&lt;/a&gt; covers the approval side. pendpost runs these checks on every lane it supports; the &lt;a href="https://pendpost.com/capabilities" rel="noopener noreferrer"&gt;capabilities page&lt;/a&gt; lists which lanes publish from the cloud and which stay local, and you can try the whole flow without real accounts in &lt;a href="https://docs.pendpost.com/mock-mode" rel="noopener noreferrer"&gt;mock mode&lt;/a&gt;.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Originally published on &lt;a href="https://pendpost.com/blog/social-media-api-posting-limits" rel="noopener noreferrer"&gt;pendpost.com&lt;/a&gt;. pendpost is a free, open-source (MIT), local-first social media planner where an AI agent drafts posts behind a human approval gate.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>api</category>
      <category>news</category>
      <category>socialmedia</category>
    </item>
  </channel>
</rss>
