AI agents

MCP server

Marsial has an MCP server so Claude, Cursor, ChatGPT and any compatible agent can publish, reply and read analytics for you, with no code.

Address: https://api.marsial.dev/mcp

claude mcp add --transport http marsial https://api.marsial.dev/mcp

Then, in Claude Code, run /mcp and pick marsial to sign in.

In .cursor/mcp.json:

{
  "mcpServers": {
    "marsial": { "url": "https://api.marsial.dev/mcp" }
  }
}

Add a remote (HTTP) MCP server with the address https://api.marsial.dev/mcp. In ChatGPT and Claude.ai it is a “custom connector”.

The first time, the agent opens the browser: you sign in to Marsial and allow it. From then on it acts on your behalf, in your workspace and with your permissions (a member cannot do what their role does not allow). Disconnect it any time in the dashboard, under Developers → AI agents.

Automation tools that only send headers (n8n, Make) can use an API key in Authorization: Bearer instead of signing in.

What it can do

Tool	For
`list_profiles`, `list_accounts`	Knowing which profiles and accounts there are.
`create_post`, `list_posts`, `get_post`, `cancel_post`, `delete_post`	Publishing, scheduling, drafting or queueing, and following them.
`list_commented_posts`, `list_comments`, `reply_to_comment`	Reading and replying to comments.
`list_conversations`, `list_messages`, `send_message`	Direct messages.
`list_reviews`, `reply_to_review`	Google Business and Facebook reviews.
`get_followers`, `get_top_posts`	Analytics.
`list_blogs`, `list_articles`, `create_article`	Articles on WordPress, Shopify, Wix and Webflow.

Documentation for the agent

All of this documentation also exists as plain text, made for language models:

/llms.txt: the index, with a summary of each page.
/llms-full.txt: the whole documentation in one file.
/llms-small.txt: a compact version, for small contexts.

Give one of them to your agent (or paste the URL into Cursor, Claude or ChatGPT) together with the OpenAPI document and it can write the integration with the real names, fields and errors.

Good practice with agents

Give it an API key from a test workspace, not production.
Have it handle the errors’ code, not the message text.
To publish on behalf of your customers, have it create one profile per customer and keep its externalId.