Your brand voice,
inside your AI assistant
Connect Brandlyre to Claude or ChatGPT and it stops writing from scratch. It reads your actual brand — voice, audience, what you have already posted — drafts posts into your calendar, and waits for your approval.
Built on the Model Context Protocol — the open standard your assistant already speaks.
The difference between an assistant that guesses and one that knows
Your assistant is already good at writing. What it has never had is your brand — and somewhere to put the result.
It writes from your brand, not a blank prompt
Before it drafts anything, your assistant reads the same brand grounding Brandlyre's own AI uses — voice, tone per platform, audience personas, objectives, content territories. You stop pasting your brand guidelines into a chat window every time.
Drafts land in your calendar, not in a chat log
Ask for a post and it appears in Brandlyre as pending approval, with per-platform copy, hashtags and media already attached. Your assistant hands back the approval link so you can open it and decide.
It can see what you have already posted
Your assistant can read recent and scheduled posts before it writes, so it does not hand you a variation of something that went out last week — the failure mode of every chat-window workflow.
Your media library is searchable by what is in the picture
Images uploaded to Brandlyre get an AI-written description, so your assistant can find "the product shot with the blue background" without you knowing the filename — then attach it to the draft.
Pick the one that matches how you work
Most people want the first one. The second exists for terminal-based tools, and it is the only route to local file upload or publishing.
Claude, ChatGPT & Claude Desktop
No setup, no token
-
1
Open your assistant's connector settings.
- Claude — Settings → Connectors → Add custom connector
- ChatGPT — Settings → Plugin → MCP → Add server
-
2
Paste this URL:
https://brandlyre.com/mcp/brandlyre -
3
Your browser opens Brandlyre to authorize the connection. Approve it and you are done.
It can never publish on its own. Connections made this way read your brand and create drafts for your approval — nothing more.
Claude Code & other MCP clients
A token you control
Create a token in Brandlyre under Profile → API & agents, then add the server:
claude mcp add --transport http brandlyre https://brandlyre.com/mcp/brandlyre \
--header "Authorization: Bearer <your-token>"Choose the access level
-
Read only — Read brands, brand voice, connected accounts and posts. Nothing can be created or changed.
-
Create drafts — Also create posts — every one lands in your approval queue — and add media to your library.
-
Publish without approval — Also approve and publish to your connected accounts without you seeing them first.
This route is also the only one that can upload a file from your own machine into the library — a browser-based assistant has no shell to read it from.
Use Brandlyre inside n8n and Zapier
The same connection works from a workflow. Point the tool's MCP client at Brandlyre and each step can list your brands, read posts and create drafts — everything in the tool list below, with the same access levels.
https://brandlyre.com/mcp/brandlyreUse a "Create drafts" token for automations. Create one in Brandlyre under Profile → API & agents. Posts your workflow creates wait in your approval queue. A "Publish without approval" token in an unattended workflow publishes with no human looking first — only use one if that is exactly what you want.
n8n
Self-hosted, version 1.104 or later
-
1
Add an MCP Client node to your workflow.
-
2
Set Server Transport to HTTP Streamable and paste the URL above into MCP Endpoint URL.
-
3
For Authentication, pick Bearer Auth and create a credential. Paste only the token into Bearer Token; n8n adds the "Bearer" prefix itself.
-
4
Choose a tool, for example create-post, and pass its arguments in JSON mode. The new post's id comes back as structuredContent.post_id for the next step.
Prefer no token? Pick MCP OAuth2 instead. n8n registers itself and opens Brandlyre so you can approve it, and a connection made this way can read and create drafts but never publish.
Zapier
MCP Client app, in beta at Zapier
-
1
In a Zap, add the MCP Client app and choose the Run Tool action.
-
2
Connect an account: paste the URL above into Server URL and set Transport to Streamable HTTP.
-
3
Answer No to OAuth and paste only the token into Bearer Token.
-
4
Pick a tool such as create-post and map its fields from the earlier steps of your Zap. Its result comes back as Structured Content fields you can use in the next steps.
Prefer no token? Answer Yes to OAuth instead. Zapier opens Brandlyre so you can approve it, and a connection made this way can read and create drafts but never publish. Run Read-Only Tool also works as a search step, for example to look up a brand or a post before acting on it.
Limits worth knowing
-
Each token can make 60 requests a minute and 2,000 a day. A tool step takes several requests, because the client opens a session for each one: in n8n it is three, so plan on about 20 steps a minute and 650 a day. Past the limit, calls fail with HTTP 429 until the window resets.
-
The limit belongs to the token, not to the workflow. Give each automation its own token, so a busy one can't use up another's budget.
-
There are no event triggers yet: a workflow can't start when a post is published. Schedule the workflow instead and read your posts with list-posts.
Nothing goes live without you
Handing an agent your social accounts should make you nervous. These are not promises in a policy document — each one is enforced in the code path.
Every draft waits for you
Posts created by an assistant are always created as pending approval. There is no argument, flag or phrasing that can skip that — it is fixed in the code path, not a default you might have left on.
Connectors can never publish
Connect through Claude, Claude Desktop or ChatGPT and the connection is capped at reading your brand and creating drafts — whichever assistant it is. Publishing requires a token you deliberately create yourself, in a terminal.
It cannot see what you did not grant
Access levels are enforced by hiding tools, not by refusing them. A read-only connection never even sees that the publishing tools exist, so there is nothing for an assistant to try and fail at.
You can always tell what came from an agent
Every post records how it was created, so agent-authored drafts are marked in your calendar. Tokens expire after a year, and revoking one disconnects that client immediately.
Thirteen tools, and you decide how many it gets
A connector-based assistant sees the first ten. The last three take a publish-level token you create yourself.
Read
6 toolsIncluded in every connection
- List your brands
- Read the brand voice grounding
- Check connected accounts
- List recent and scheduled posts
- Read one post in full
- Search the media library
Create drafts
4 toolsConnectors and "Create drafts" tokens
- Create a post (pending approval)
- Edit a draft or pending post
- Upload a local file to the library
- Import an image from a URL
Publish
3 toolsOnly a publish-level token, created by you
- Schedule a post
- Approve a pending post
- Queue a post to publish now
Questions about connecting your assistant
Does this work with ChatGPT, or only Claude?
Both. Brandlyre is a remote MCP server, so any assistant that supports MCP connectors can use it — Claude, Claude Desktop and ChatGPT have all been connected. The setup is the same in each: add Brandlyre as a custom connector and authorize it in your browser. Every connector-based connection gets the same access: read your brand and create drafts, never publish.
Do I need to be a developer?
No. The connector route is a URL you paste into your assistant's settings and a button you click in the browser — there is no token to copy, no config file to edit, and no terminal. The terminal route exists for Claude Code and similar CLI tools, and it is the only way to get local file upload or publishing, but most people never need it.
Can my assistant post to my social accounts without me seeing it first?
Not over a connector. Connections made from Claude, Claude Desktop or ChatGPT are limited to reading your brand and creating drafts, and the publishing tools are not visible to them at all. The only way an agent can publish unattended is if you deliberately create a publish-level token in Brandlyre and choose that access level yourself — the option exists, clearly labelled, for people who want it.
Which platforms can it post to?
Only the ones you have already connected to that brand in Brandlyre. Your assistant checks your connected accounts before drafting and will tell you if a platform is missing rather than guessing.
Does using it cost credits?
Working through your assistant does not spend credits — the assistant writes the copy itself, so Brandlyre's own AI is not doing the work. Credits apply where they always do: when Brandlyre generates something for you. The one thing to know is that images added to your library get an automatic AI description, which costs a single credit and can be turned off per upload.
Can I use Brandlyre from Zapier or n8n?
Yes. Both have an MCP client that connects to the same URL as the assistants, with a token you create. Use a "Create drafts" token for automations, so the posts a workflow creates wait for your approval. There are no event triggers yet, so a workflow can't start when a post is published.
How do I disconnect?
Revoke the connection from API & agents in your Brandlyre profile, or remove the connector from your assistant's own settings. Revoking takes effect immediately — any client using that connection stops working straight away. Tokens also expire on their own after a year.
Do you train AI models on my content?
No. Your brand profile, posts and media are used to generate your content and nothing else. Copy and image generation run on Google Gemini by default, with OpenAI as an opt-in fallback, and we do not train models on your content.
Give your assistant something to work from
Set up your brand in Brandlyre, connect it to Claude or ChatGPT, and let it draft. You still approve everything.
No credit card required. Set up in under 10 minutes.