---
title: "Claude Code integration with Bolta"
description: "Connect Bolta to Claude Code with /plugin marketplace add boltaai/bolta-skills, then draft LinkedIn and X posts from your changelog. Drafts wait for your approval."
canonical_url: "https://bolta.ai/integrations/claude-code"
markdown_url: "https://bolta.ai/integrations/claude-code.md"
last_updated: "2026-11-21"
content_type: "integration"
publisher: "Bolta"
---

# Claude Code integration with Bolta

Source: https://bolta.ai/integrations/claude-code
Last updated: 2026-11-21

Connect Bolta to Claude Code with /plugin marketplace add boltaai/bolta-skills, then draft LinkedIn and X posts from your changelog. Drafts wait for your approval.

## Summary

Turn your repo's release notes into social drafts, from the terminal

## Who this is for

- Developers
- Automation teams
- AI-assisted marketing teams

## Core capabilities

- Turn the newest CHANGELOG.md entry into a LinkedIn and an X draft
- Draft launch posts from a README section before you push a release
- Summarise what a merged pull request changed in plain words for customers
- Draft a week of build-in-public posts in your own voice between coding sessions
- Review what Bolta has queued without leaving the terminal
- Set up a weekly routine that drafts posts from your product updates for your review

## Limitations and boundaries

- Confirmed in Claude Code's official docs: remote HTTP MCP servers with browser sign-in through /mcp, skills as SKILL.md folders, and plugin marketplaces added from a GitHub repo.
- Bolta's own plugin is the most tested path. A full real-world run on every setup is still beta (plugin v0.5.3).
- Not confirmed here: the exact minimum Claude Code version for plugin marketplaces (the Bolta plugin README names 2.1.295 or later only as an example), and running Bolta end to end from claude -p in scripts. Claude Code's docs say headless runs cannot do the browser sign-in. The API key header route follows Claude Code's documented --header option but was not run end to end.
- Integration availability depends on current APIs, authentication, scopes, and account permissions.
- Generated content can be incorrect and should be reviewed before publication.
- Availability depends on the current Bolta plan, connected network, account permissions, and integration coverage.
- Current prices and plan limits must be verified on the Bolta pricing page.

## Start Claude Code and find its prompt

Open a terminal and start Claude Code. The next two commands are typed into the Claude Code prompt, not into the terminal itself.

## Add the Bolta marketplace

Run /plugin marketplace add boltaai/bolta-skills. A marketplace is a list of plugins Claude Code can install, and boltaai/bolta-skills is the GitHub repository that holds it. Run /plugin and look for bolta-marketplace to check it worked.

## Install the Bolta plugin

Run /plugin install bolta@bolta-marketplace and pick user scope unless you know you want another. The plugin brings the connection to https://mcp.bolta.ai/mcp and the Bolta skill together. If the summary says to run /reload-plugins, run it.

## Sign in from /mcp

Run /mcp, select Bolta and choose sign in. Your browser opens on bolta.ai. Sign in, pick your organization and tick the brands this connection may reach. There is no key to copy.

## Paste the address if the browser errors

If the browser shows a connection error at the end, copy the full address from the address bar and paste it into the prompt Claude Code shows. Then go back to the terminal.

## Confirm Bolta is connected

Run /mcp again. Bolta should show as connected. If it still asks you to sign in, repeat the sign-in step and finish the browser page before returning.

## Say set me up

Type set me up in the chat, or run /bolta:bolta. Claude reads your brands and asks for your site, your voice or which accounts to connect. Drafts save freely. Nothing goes out until you say yes.

## Check it works without changing anything

Ask: Which Bolta brands can you see, and which social accounts are connected to each? Do not change anything. Claude should list the brands you ticked and their networks, and draft nothing.

## Already connected Bolta yourself? Add only the skill

If you added Bolta earlier with claude mcp add or a connector, skip the plugin. Run npx skills add boltaai/bolta-skills --skill bolta in a terminal to add just the skill. Optionally run /plugin install bolta-checks@bolta-marketplace for a check before anything public.

## Turn your repo into posts

Claude Code already reads your files, so the source for a post is often sitting in the project. Point it at a changelog, a README section, release notes or a diff summary and ask for drafts. It uses your Bolta voice and your posting rules, saves the drafts to Bolta and waits. Claude Code reads the files itself, so there is no copy and paste between your editor and a scheduling tool. Name the file, for example CHANGELOG.md, so it does not guess. Say which brand when you ticked more than one. Ask for drafts only. Approval stays a separate yes.

## Plugin, skill or raw MCP: avoid two Bolta entries

The plugin installs the connection and the skill. If you already added Bolta yourself, installing the plugin on top leaves two Bolta entries on one organization. In that case add only the skill with npx skills add boltaai/bolta-skills --skill bolta. The optional bolta-checks plugin shows a check before anything public or final, such as posting, replying, approving, scheduling, running a routine now or starting a campaign. Saving drafts needs no check. On Windows it needs Git for Windows, because the checks use sh and awk and every Bolta save is blocked without them.

## API keys, scripts and team files

Claude Code's docs describe adding a static key with --header "Authorization: Bearer \<token\>", or putting it under headers in .mcp.json with ${VAR} expansion. Bolta supports API keys, so that route should work, but it was not run end to end for this page. Prefer sign-in. Posts from a sign-in are held as AI drafts for review, and a pasted Bolta API key skips that hold. For scripts, Claude Code's docs say claude -p runs cannot perform the browser sign-in, so sign in first from an interactive session with /mcp. We have not run Bolta end to end from claude -p or CI, so test it yourself and keep a person in the loop for approval.

## Scopes: user, project and local

Claude Code adds servers at local scope by default, which is just you in the current project. User scope follows you across all your projects. Project scope writes a .mcp.json file in the project that can be committed so a team shares a setup, and Claude Code asks each person to approve it in an interactive session. Plugins install at user, project or local scope too. Each person still signs in to Bolta as themselves, and the sign-in follows their own Bolta role. One connection is one organization. Agencies tick their client brands in the same connection. Several organizations need one connection each.

## The bolta-checks plugin: a pause before anything public

Saving a draft is harmless. Posting, replying, approving or scheduling is not, so some people want a second stop in the terminal. bolta-checks is a separate plugin in the same marketplace. Install it with /plugin install bolta-checks@bolta-marketplace. It shows a check before the public or final Bolta actions: posting, replying, approving, scheduling, running a routine now or starting a campaign. Plain draft saves pass through without a prompt. It is optional and it sits on top of Bolta's own rules, which already refuse an action your role does not allow. If you are on Windows, install Git for Windows first. The checks need sh and awk, and without them every Bolta save is blocked. Install it only after the main plugin works. Remove it with the plugin manager if saves start failing on Windows. It adds a pause. It does not change who is allowed to approve.

## Build in public without leaving your editor session

Most developers skip posting because writing it means switching context. Here the context is already in front of the agent. Between tasks, ask Claude Code to describe what you shipped since your last post and draft it in your voice. It can read the commit messages, the changelog and the files you touched. It does not know why a change mattered to a customer unless you tell it, so add one line of intent to the prompt. Ask for two or three angles on the same change, for example the problem, the fix and a lesson, and pick one. Saving drafts is free, so keep all of them until you choose and delete the angles you do not use. Give the intent in one sentence: who benefits and why. Ask it to flag any claim it cannot back from the repo. Name the brand and the network so the right rules apply.

## A release-day habit that takes five minutes

This is a suggested sequence, not a built-in feature. Every step stays a draft until you say yes. Before you tag the release, ask Claude Code to read the top CHANGELOG.md entry and save one draft per network you use. Read each draft in the terminal. If a draft overstates what shipped, say what to cut and ask again. When the release is live, come back and say yes to the draft you like. Claude should show the exact text and the account first. If you would rather a teammate signs off on it, send it for review instead of approving it yourself. Draft before the release, approve after you have checked the release is real. Keep screenshots as public links, since media goes in by link. Never let a draft name a feature the changelog does not mention.

## What is Claude Code, and how is it different from Claude or Claude Cowork?

Claude Code is Anthropic's coding agent that runs in your terminal and reads your project files. Claude is the chat app, and Cowork and Duet have their own Bolta pages. Bolta behaves the same in all of them: drafts are free and nothing publishes without your yes.

## How do I connect Bolta to Claude Code?

Run /plugin marketplace add boltaai/bolta-skills, then /plugin install bolta@bolta-marketplace, then /mcp and sign in. Your browser opens on bolta.ai and you tick the brands Claude Code may reach.

## Do I need an API key?

No. Signing in from /mcp is the recommended route and needs no key. Posts from a sign-in are held as AI drafts on your review list, while a pasted API key skips that hold.

## Plugin, skill or plain MCP: which should I use?

Use the plugin if you have not connected Bolta before, since it brings the connection and the skill together. If Bolta is already connected, add only the skill. Installing both leaves two entries on one organization.

## Can Claude Code publish without my approval?

Not by design. Bolta's rule is that nothing is approved, scheduled or published until you say yes, and Claude should show you the exact text first. Posts from a sign-in are held as AI drafts on your review list, and Bolta refuses an action your role or settings do not allow.

## Can I use Bolta from a script or CI with Claude Code?

Claude Code documents a --header option and .mcp.json headers for static keys, and Bolta supports API keys. Its docs also say claude -p runs cannot do the browser sign-in, so you would sign in first from an interactive session. We have not run Bolta end to end from claude -p or CI, so test it yourself and prefer sign-in for anything a person reviews.

## Claude Code does not see the Bolta tools. What should I check?

Run /reload-plugins, then /mcp, and make sure Bolta shows as connected rather than needing sign-in. If you see two Bolta entries, remove one.

## Which brands can Claude Code reach?

Only the brands you ticked when you signed in. If you ticked more than one, name the brand in your request. To add one later, run /mcp, pick Bolta and sign in again.

## Which social platforms can it post to?

Any network connected to your Bolta brand, including LinkedIn, X, Instagram, Facebook, Threads and Bluesky. It cannot post to an account you have not connected in Bolta. Instagram and Pinterest need an image or video, and TikTok and YouTube need a video.

## What can it not do?

It cannot do more than your Bolta role allows. A viewer connection is read-only and a client reviewer cannot connect at all. Media goes in as a public link, since uploading from chat is not available.

## Does the bolta-checks plugin work on Windows?

It needs Git for Windows installed first, because the checks call sh and awk. Without them every Bolta save is blocked, so install Git for Windows and restart Claude Code, or uninstall bolta-checks.

## Can my whole team share one Bolta setup through the repo?

Claude Code's project scope writes a .mcp.json file you can commit, and each person approves it and signs in as themselves. Their Bolta role decides what they can do, so a shared file never shares one person's permissions.

## The sign-in page opens again and again, or Bolta stays on not signed in.

Run /mcp, pick Bolta and sign in once more, and finish the browser page all the way before going back to the terminal. If the browser shows a connection error, paste the full address from its address bar into the prompt Claude Code shows.

## Claude Code does not see the Bolta tools.

Run /reload-plugins if the install summary asked for it, then run /mcp and confirm Bolta shows as connected. A plugin installed from the terminal with claude plugin install loads the next time you start Claude Code.

## I see two Bolta entries in /mcp.

You installed the plugin and also added Bolta yourself. Keep one and remove the other. If you keep your own connection, add only the skill with npx skills add boltaai/bolta-skills --skill bolta.

## A brand I expect is missing.

It was not ticked at sign-in, or you added it afterwards. Run /mcp, pick Bolta, sign in again and tick it. If you ticked several brands, name the one you mean in your request.

## Claude says the sign-in expired or you are not authenticated.

Run /mcp, pick Bolta and sign in again. Sign-ins last 60 days and normally renew themselves. Claude should not retry in a loop.

## Every Bolta save is blocked on Windows.

You have bolta-checks installed and no sh or awk. Install Git for Windows and restart Claude Code, or uninstall bolta-checks.

## Claude Code is not Claude, Claude Cowork or Claude Duet

Claude Code works in a terminal, next to your files and git history, so it suits developer-led posting. Claude is the chat app, and Cowork and Duet have their own pages with their own setup. The Bolta account and approval rules are the same in all of them.

## Relevant links

- [Related page: /integrations](https://bolta.ai/integrations)
- [Related page: /integrations/claude](https://bolta.ai/integrations/claude)
- [Related page: /integrations/claude-cowork](https://bolta.ai/integrations/claude-cowork)
- [Related page: /integrations/claude-duet](https://bolta.ai/integrations/claude-duet)
- [Related page: /integrations/codex](https://bolta.ai/integrations/codex)
- [Related page: /integrations/cursor](https://bolta.ai/integrations/cursor)
- [Bolta pricing](https://bolta.ai/pricing)

## Supporting sources

- [Claude Code official website](https://claude.com/product/claude-code)
