OutstandOutstand
Engineering

Social media MCP servers: which ones actually publish

Most social media MCP servers publish, but not all of them. A dated inventory of which servers post, which AI clients each supports, and what plan tier you need.

Social media MCP servers: which ones actually publish

Most social media MCP servers can publish. The ones that cannot are not the ones people expect. Bright Data's social media MCP server extracts public social data and does not post at all. Ayrshare ships two servers and one of them is documentation only, so connecting the wrong one gives your agent a search tool for API docs instead of a publisher. Sprout Social's MCP server is announced as a way to pull social data into AI workflows, and its public materials do not describe publishing.

Below is an inventory of the social media MCP servers we could verify against each vendor's own documentation on 30 August 2026: which ones publish, which AI clients each supports, and what you need to pay to get access. Every list we found tells you a server exists. Almost none tell you whether you can actually use it, which is the part that decides your build.

We run one of these servers, so treat the recommendation accordingly. The last section says plainly when Outstand is the wrong choice.

What is a social media MCP server?

An MCP server exposes tools to an AI client over the Model Context Protocol, so an assistant such as Claude or ChatGPT can call them without you writing integration code. A social media MCP server exposes social media operations as those tools.

The category splits in two, and the split is the most common source of wasted setup time:

  • Publishing servers let an agent create, schedule and analyse posts on accounts you own. This is what most people mean when they search for one.
  • Data servers let an agent read public social content: posts, profiles, comments, trends. They are built for research and monitoring, and they do not post.

A third case catches people out. Some vendors ship more than one MCP server. Ayrshare ships an action server that publishes and a separate documentation server that only searches its API docs. Hootsuite ships four servers split by job. Connecting the wrong one does not produce an error. Your agent simply does not have a posting tool, and you work out why later.

Which social media MCP servers exist right now?

Verified against each vendor's own live documentation on 30 August 2026. Where a vendor does not publish a number, this says so rather than guessing. This is a fast moving category and we re-check this list quarterly.

  • Outstand. First-party. Publishes. 28 tools across 12 platforms, at mcp.outstand.so/mcp. One-click OAuth or an API key as a Bearer token. Any client that supports remote MCP servers.
  • Ayrshare. First-party, two servers. The action server publishes and exposes 27 tools at api.ayrshare.com/mcp over Streamable HTTP. The documentation server is read-only and searches its API docs.
  • Buffer. First-party. Publishes into your Buffer queue rather than straight to the network. Tool count not published. Available on every plan including Free.
  • Hootsuite. First-party, four servers: Perch for publishing and planning, Nest for the inbox, Lumen for listening, Parliament for advocacy. Perch is the one that posts. Tool counts and plan tiers not published. See hootsuite.com/integrations/mcp.
  • Vista Social. First-party. Publishes. More than 50 tools spanning publishing, analytics, inbox and listening. Included with the Advanced plan and higher.
  • Postiz. First-party. Publishes. 11 tools, including schedulePostTool for scheduling, drafting or immediate publishing. Available on cloud at api.postiz.com/mcp and on self-hosted installs.
  • Sprout Social. First-party. Its November 2025 announcement describes a remote MCP server that lets customers "embed social data directly into day-to-day AI workflows", starting with ChatGPT. It does not describe publishing. Its ChatGPT connection page describes summarising and analysing the social data already in Sprout rather than posting to it.
  • Bright Data. First-party. Does not publish. It is built to extract and analyse public social data at scale: posts, profiles, comments, reactions.
  • tayler-id/social-media-mcp. Community project, not a vendor product. Publishes to X, Mastodon and LinkedIn through three tools. It is the top organic result for this search and it has one commit and 22 stars.

Which ones cannot actually post?

Three cases, and none of them is obvious from a directory listing.

  • Bright Data is extraction only. This is not a shortcoming, it is the product. If your job is research, monitoring or competitor analysis, this is the right tool and a publishing server is not.
  • Ayrshare's documentation server is read-only by design. Ayrshare publishes through its action server. The documentation server exists so a coding agent generates correct integration code. Both are legitimate, and they are easy to confuse when you are copying a URL into a config file.
  • Sprout Social's is announced as a data integration. The company's own release positions it around analysing campaign performance and identifying top-performing content. Publishing is not mentioned. Several third-party MCP servers for Sprout Social do exist and some of them do post, so a search result claiming Sprout publishing over MCP may be describing a community project rather than the official server.

Everything else in the list above publishes: Outstand, Ayrshare's action server, Buffer, Hootsuite's Perch, Vista Social, Postiz and the community repo.

Which AI clients does each one support?

This is the question that decides whether a server is usable for you, and it is the one most lists skip. "Supports MCP" is not the same as "supports your client".

  • Sprout Social: ChatGPT is the named starting point in its announcement. If your team works in Claude or Cursor, check before you plan around it.
  • Hootsuite: ChatGPT, Claude, Microsoft Copilot and Google Gemini.
  • Buffer: Claude, ChatGPT, Cursor, Raycast, Perplexity and others.
  • Vista Social: Claude, ChatGPT via custom connectors, Cursor, Gemini, Windsurf and others.
  • Outstand: any client that supports remote MCP servers. Our setup guide covers ChatGPT, Claude, Claude Code, Cursor and Windsurf explicitly.
  • Ayrshare and Postiz: any MCP client. Ayrshare additionally ships a Claude Code plugin.

A practical note on transport. Remote servers that speak Streamable HTTP with OAuth are the ones that connect cleanly to hosted clients like ChatGPT and Claude without a local bridge. Servers you clone and run yourself usually need a stdio bridge and a config file, which is fine on a laptop and awkward in CI.

What does it cost to get MCP access?

Plan gating is real and it is almost never mentioned in the listicles. What each vendor states publicly:

  • Buffer: every plan, including Free.
  • Vista Social: Advanced plan and higher.
  • Postiz: cloud, or self-host it yourself.
  • Outstand: included with an Outstand account, and signing up is free.
  • Bright Data: paid, priced as data extraction.
  • Hootsuite, Sprout Social and Ayrshare: not stated on their public MCP pages. Ask their sales or support teams rather than assuming it is included in your current plan.
  • tayler-id/social-media-mcp: free, and you host it.

How do you tell a maintained server from an abandoned one?

The top organic result for "social media mcp" today is a personal GitHub repository with one commit and 22 stars. It works, it exposes create_post, get_trending_topics and research_topic, and it covers X, Mastodon and LinkedIn. It is also someone's side project, and it ranks above every vendor because of how search works, not because of what it is.

That is not a criticism of the author. It is a warning about how you choose. Check four things:

  • Commit history. One commit means the code has never been revised after first publication. Social platform APIs change constantly and a server that is not maintained breaks quietly.
  • Who answers when it breaks. A vendor server has a support channel. A community repo has an issue tracker and the maintainer's spare time.
  • Whether tokens are handled for you. OAuth token refresh across a dozen networks is most of the real work. TikTok access tokens, for example, last about 24 hours.
  • Whether the server is versioned against the MCP spec. The protocol is still moving. A server that names the revision it implements is telling you someone is tracking it.

What breaks when an agent posts to social media?

This is the section most posts on this topic leave out, and it is the part that decides whether your integration survives contact with production. The failure modes below are how our own pipeline behaves, documented in our public API docs. Other servers differ, and you should ask them the same questions.

  • Publishing is asynchronous. A successful API call means the post was accepted, not that it went live. If your agent reports success on the response alone, it will lie to you.
  • Multi-network posts are not all-or-nothing. Each account is sent independently and a failure on one does not abort the others. A post can be published on LinkedIn and failed on Facebook at the same time. Always inspect per-account status rather than treating the post as a single boolean.
  • Some failures are terminal and must not be retried. An expired token, rejected content or invalid media is recorded as failed and is deliberately not retried, because a blind retry is how you get duplicate posts. Transient network failures are retried for you with exponential backoff.
  • Rate limits and billing errors look similar and are not. A 429 is retryable and carries a Retry-After header to honour. A 402 is not retryable and will keep failing until the underlying account problem is fixed.
  • Deletion is not universally supported. TikTok and Instagram do not support deleting a published post through their APIs. An agent that can post to those networks still cannot take the post back down for you. Plan for a human to do it.
  • Media uploads are multi-step. Instagram requires container creation then polling, and X uses chunked uploads. An agent that treats media as a single call will fail intermittently on large files.

The general point holds regardless of which server you pick. Giving an agent a posting tool is easy. Giving it an accurate picture of what happened after it posted is the actual engineering problem, and it is worth asking any vendor how their tools report partial failure before you build on them.

When should you not use Outstand?

We publish through our own MCP server, including the distribution for this post, so we are not neutral. Here is where we are the wrong answer:

  • You only need to read social data. Use Bright Data. We are a publishing API and we do not scrape public content, run a firehose or do sentiment monitoring.
  • You already pay for Buffer, Hootsuite, Sprout Social or Vista Social. Their MCP servers are first-party and included in what you are already buying. Adding us is a second bill for a job your current tool now does. If their client support and plan tier work for you, use theirs.
  • You want free and self-hosted. Postiz is a genuine option, it publishes, and you can run it yourself. We are hosted and we are not free forever.
  • You only ever post to one network. Use that network's own API. An abstraction layer over a single platform costs you a dependency and buys you very little.

Where we do fit: you are wiring an agent into several networks at once, you need one interface across all of them, and you need per-account visibility into what published and what failed. That is the case we built for. The tool reference lists all 28 tools, and the setup guide has per-client instructions.