X (Twitter) posts with MCP Server and Bolta
One address for any MCP client: https://mcp.bolta.ai/mcp. Parts of the MCP Server route are not confirmed yet, and the note below says which. Nothing goes live on X until you say yes.
Partly confirmed, see what is and is notChecked 2026-11-21
The same address works in every app. Where you paste it depends on the app.
Draft an X post of 280 characters or fewer announcing our new feature, and show me the count
Illustration of the reply: Bolta returns a draft showing the connected account and the scheduled time, with Approve, Edit and Reject actions.
What is and is not confirmed
Read this before you set up MCP Server
- Confirmed: the address https://mcp.bolta.ai/mcp answers over HTTP and asks for a Bearer token, pointing to OAuth sign-in details. Bolta's own docs describe the browser sign-in and the Bearer API key route.
- Confirmed from the MCP specification: remote servers can ask for OAuth sign-in or accept a bearer token over HTTP.
- Not confirmed: nobody has run a full draft, review and schedule in each MCP client. Bolta's skills repository lists every client except the Claude Code plugin as not tested, and the Claude Code row runs its behavioural checks against a mock Bolta.
- Not confirmed: whether your app's settings screen has a box for an Authorization header. The protocol allows it, but apps differ.
- Menu names and config file locations change between app versions. Use your app's own MCP docs for where to click, and keep the address the same.
What Bolta does on X
Bolta reads the limit for each connected X account, so a Premium account and a free account get different ceilings. Text that is too long does not get trimmed on X: the post fails, so the draft has to fit first.
- Single posts and threads. In a thread, Bolta posts every part after the first as a reply to the one before.
- Images. A post needs text: an image with no text is not supported on X.
X limits and format rules
Network rules last checked 2026-07-28. Bolta's posting rules for your workspace (get-posting-rules) win if they differ from this page.
Not promised for X from MCP Server: Polls and Spaces announcements. Neither is confirmed for Bolta. How many hashtags work best. Bolta checks none, and this page makes no claim about it.
| What to know | X's own rule | What Bolta does |
|---|---|---|
| Post length | Accounts without a paid subscription are limited to 280 characters. X Premium subscribers can publish up to 25,000. | Bolta reads the limit per account. Longer text makes the post fail, so shorten it first. The 25,000 limit is a ceiling for Premium accounts, and X can still refuse a long post. |
| Links and media | A pasted link counts as 23 characters however long it is. Photos, videos and GIFs count as zero characters. | Bolta does not check how many hashtags you use or their format. They are kept as typed. |
| Threads | Posts can be chained into a thread, and each post carries its own separate character limit. | Each X thread part after the first is posted as a reply to the one before it. |
| Who can reply | Replies can be limited to everyone, accounts you follow, verified accounts, or only the accounts you mention. It is chosen before you post. | The default is that anyone can reply unless you limit it. |
| Video | Not stated on this page. | Video on X is checked on X's side for duration, format and posting eligibility, so a successful upload does not guarantee publication. Ask for the posting rules before you attach one. |
MCP Server jobs for X, and how to write them
- Check what is waiting for approval without opening Bolta
- Set up a repeating routine from chat, with the plan shown first
- Write to 280 unless the account has X Premium. Check the count, because a post over the limit fails instead of being cut.
Example prompts for X
Things to type in MCP Server. Each one asks for a draft and shows you the text first.
Draft an X post of 280 characters or fewer announcing our new feature, and show me the count
Which Bolta brands can you see? List them only.
Write a five-part X thread about how we handle onboarding, with every part inside the limit, and show me all five before saving
Save a draft post about our newest feature in my brand voice
Bolta saves a draft, you read the exact text, you say yes
- 01
Bolta saves a draft
MCP Server asks Bolta for the X post in your brand voice (save-post). A saved draft is never public.
- 02
You read the exact text
On X the exact text includes the character count with links counted at 23, and, for a thread, every part in order.
- 03
You say yes
Only then does Bolta schedule or publish (publish-post), in your workspace's timezone. Safe Mode, client sign-off and brand approval rules still apply.
Using MCP Server for X
Parts of this route are not confirmed. The note above says which.
For X, MCP Server, from Open protocol (Anthropic started it), reaches Bolta through Remote server address, Streamable HTTP. Sign-in: Sign in (OAuth), or a Bearer API key. The X account itself is connected inside Bolta, not in MCP Server.
The Bolta MCP server address works in any app that adds remote MCP servers. This page is for apps without a page of their own, and an X draft works the same way once your app is connected.
An illustration from the MCP Server guide, not real output. You ask: Show me what is on my Bolta review list. The reply goes along these lines: Five items are waiting: three drafts from a routine and two from this chat. I can read out the full text of any of them. I will not approve any until you tell me which ones. Nothing approved.
An illustration from the MCP Server guide, not real output. You ask: Plan a weekly Friday post for Harbor Coffee and show me the plan first. The reply goes along these lines: Proposed plan: a draft each Friday at 9:00, brand time, sent to review before anything goes out. Bolta would run it, so it repeats with this app closed. If you agree I save it switched off, and turning it on is a separate yes from you. Plan shown. Waits for your yes.
Your role and the key's permissions decide what each call may do. Anything that changes what goes out, such as approving, scheduling or publishing, needs your yes to that task, and Safe Mode applies on top.
A skill is a folder of written instructions your app can load so it uses Bolta's tools in a sensible order. It is optional. The connection works without it, and the skill only helps in apps that read a skills folder.
The command is npx skills add boltaai/bolta-skills --skill bolta. By hand, copy skills/bolta from the repository, and keep the folder name.
- Review and approval. See what is waiting and send a draft to the team review list. Approval is tied to the exact text, so a changed post needs a fresh yes. Bolta's skills README says batch approval and rejecting with a reason reach Claude and Cursor today, and other apps fall back to one approval at a time or a link to Bolta.
- Voice and brand. Read the voice the AI writes in, add a voice profile, and read or change the business details it works from.
- The AI reads your voice. get-voice-context hands the AI the voice and business details you set up in Bolta, so the draft is written from those and not from the AI's own default style.
- You decide. The AI shows you the exact text. publish-post is called only after you say yes to that post, and Bolta's own checks can still refuse it.
- Setup: Find where your app adds MCP servers. Look for MCP or Connectors in the app's settings, or for a JSON config file. If you find neither, the app may not support remote MCP servers yet, so check its release notes or MCP docs.
- Setup: Ask for a first draft. Ask for three draft posts for next week. They save as drafts you can read and edit, and none of them is approved, scheduled or published until you say yes.
- Claude Code plugin. Not part of this generic route. Claude Code has its own page with its own steps.
- A routine saved in Bolta repeats on its schedule even with your app closed. It is created from chat, but Bolta runs it.
- API key (Bearer header). Choose this when the app only accepts headers and has no connect button. A key you create acts as you, so it is not held back by the review list the way the sign-in route is. Safe Mode and a client's required sign-off still apply. Tick only what the job needs, for example posts:write to save posts.
Connect MCP Server to Bolta
The same address works in every app. Where you paste it depends on the app.
X questions, answered for MCP Server
- What happens if MCP Server writes an X post that is too long?
- The post fails. Bolta does not trim X text, so MCP Server has to shorten it first. The limit is 280 characters for accounts without a paid subscription, and X Premium accounts can post up to 25,000, though X can still refuse a long post.
- Can MCP Server write an X thread?
- Yes. Ask for several parts, and Bolta posts each part after the first as a reply to the one before it, so they read as one thread. Each part carries its own limit. You read all of the parts in MCP Server and say yes before any of them goes out.
- What is Bolta's MCP server address?
- It is https://mcp.bolta.ai/mcp. It is a remote server over Streamable HTTP. If your app asks for a type, choose HTTP.
- How is the MCP server different from the Bolta REST API or the BoltaClaw CLI?
- The MCP server lets an AI app call Bolta on your behalf. The REST API is for code you write yourself. BoltaClaw is a separate runtime with its own page.
- How long does a sign-in last?
- About 60 days per Bolta's guides, and they usually renew without you doing anything. If a call says Invalid or expired, disconnect Bolta in your app and sign in again.
- The AI asks which brand to use, or a call asks for a workspace_id.
- You have several brands on one connection. Name the brand in your prompt, for example: for Acme, show my review list. Bolta then passes that brand's workspace_id on each call.
- Does MCP Server publish X posts without my yes?
- No. Nothing is approved, scheduled or published until you say yes to that X post. Whether MCP Server completes the connection is not fully confirmed, so read the note above first.
More from MCP Server
MCP Server can work with the other networks Bolta publishes to. Each page has its own limits and review notes.
X from other AI agents
The same X limits apply whichever agent asks. The connection and the setup are what change.
Try an X draft from MCP Server
Connect MCP Server, ask for an X draft and read the exact text. Nothing goes live until you say yes.