Run your marketing from Claude Code with the Oetzi MCP server

Claude Code is Anthropic's coding agent for the terminal. It reads your project, runs commands, and calls MCP servers for anything outside the repo. Point it at Oetzi and the agent that ships your product can also do your marketing: find the conversations worth answering, draft the reply in your voice, check whether the AI engines name you, publish the post. Nothing paid runs unless you approve it.

What is Claude Code, and why give it a marketing server?

Claude Code works from the tools you hand it, and Anthropic documents the install and the MCP command in the Claude Code overview. On its own the agent has no idea what your market said this week. The Oetzi MCP server hands it a workspace: listening signals across seven platforms, a scored inbox, your voice doc, your posting plan, the people graph, and the outcomes ledger.

What can a founder do from Claude Code?

Anything you can do in the portal, because every surface renders from one registry of capabilities. A session usually looks like this:

• Ask which conversations deserve an answer this morning. Every match is scored 0 to 100 for buying intent with the reason written out, so you get three threads instead of forty.
• Draft two reply variants in your voice for any conversation, held for your approval. Nothing goes out on its own.
• Run a share of model check across ChatGPT, Perplexity, Gemini, Grok, and Google AI, and read back who got named instead of you.
• Draft today's post on the cadence your posting plan set, then publish it through Postiz once you say yes.
• Pull the people who engaged with a post and record the outreach so nobody hears from you twice.

How do I connect Claude Code to Oetzi?

Create a workspace API key in Settings, then add the server as an HTTP endpoint with that key as a bearer token. The command and the tool list live in the MCP docs. One key covers one workspace.

Oetzi also ships a SKILL.md for Claude Code that teaches the agent the capabilities, the credit rules, and the approval gate, so "check what people said about us this week and draft replies" lands on the right tools. For cron jobs, use the CLI.

What stops an agent from spending your credits?

Two gates, both enforced on the server. Keys carry scopes: read, write, spend, and post. A read key sees the inbox and the ledger and nothing else. A key without post scope cannot publish.

Then every paid call returns a quote first. The agent sees the price in credits, and the call executes only when it is resent with approved set to true and a max_credits cap. Work over the cap never starts. Reading is free, so an agent watching your inbox costs nothing. Long work returns a job it can wait on or hear about by webhook.

Questions

Can Claude Code post to social media?

Yes, when the key carries post scope and you approve the draft. Claude Code writes the post in your voice, the capability returns a quote, and publishing runs only after approval. Posts go out through your own Postiz account. Publishing itself is free, and a drafting run is one credit.

Is the marketing skill for Claude Code free?

Yes. The SKILL.md, the MCP server, the REST API, the CLI, webhooks, and events all cost nothing. You pay only for metered work such as sweeps, drafts, renders, and checks. A free account starts with 10 credits and no card, and they never expire.

Does the agent see the same data as the portal?

Yes. One registry of capabilities sits behind both, so the agent reads the same scored inbox, voice doc, and outcomes ledger you do. Ratings and statuses are shared, so a conversation the agent marks contacted reads that way in the portal too.