Oetzi
Free scanStart for free

For agents

The same growth hacker,
reporting to your agent.

Everything on the hire page is also 145 capabilities over MCP, REST and the CLI. One envelope, the same names everywhere. It quotes before it spends and logs what it sent.

Read the agent docsGet a workspace key
Claude CodeCursorClaude Desktop
one session
MCPRESTCLISKILL.md
$ claude mcp add oetzi https://app.oetzi.ai/api/mcp

> capabilities_search "hot matches this week"
  get_leads · list_people · draft_posts

> get_leads band=hot limit=3
  3 matches, top score 91: "anyone found a tool that…"

> draft_posts platform=linkedin topic_id=t_8f2
  quote: 1 credit. re-run with approved=true

> draft_posts platform=linkedin topic_id=t_8f2 approved=true max_credits=1
  drafted, waiting for a tap in the workspace

What your agent gets

Six things it cannot see on its own.

An agent handed a list messages strangers. Handed a scored thread and the person behind it, it shows up as a peer. Oetzi is that feed.

The scored inbox

read

Every conversation on seven platforms, scored 0 to 100 for buying intent, with a drafted reply. get_leads, get_lead, mark_lead.

The people graph

readspend

Who is behind the thread, who engages with your competitors, who you already talked to. list_people, pull_engagers, record_outreach.

Share of model

spend

Whether ChatGPT, Perplexity, Gemini and Grok name you for the buying questions, and who they name instead. run_geo_check, get_share_of_model.

The knowledge store

readwrite

What shipped, what customers said, what the research pulled in. search_knowledge, add_knowledge, set_feed. Every draft cites it.

Posting and video

spendpost

Drafts in the workspace voice, scheduled through Postiz, short video with captions. draft_posts, publish_post, render_content.

The trust ladder

write

A rung per workspace per kind of work. Your agent reads it, proposes a rung, and never moves one up without a yes. get_trust, set_trust, hold_post.

How it works

Connect in one command.

  1. 01

    Connect

    One MCP command, or a REST key, or the CLI. Keys carry scopes: read, write, spend, post. Give the agent only the scopes the job needs.

  2. 02

    Discover

    capabilities_search finds the right call by intent, capabilities_schema returns its exact input. 145 capabilities, one envelope, same names everywhere.

  3. 03

    Quote, then spend

    Any paid call without approved returns a dry run: what it would do and what it would cost. Re-run with approved and max_credits and it charges inside that ceiling.

  4. 04

    Wait, then report back

    Long work returns a job_id for wait_job. Every change lands in an event log: long-poll wait_events or take a signed webhook. record_outreach closes the loop so nobody gets messaged twice.

Guardrails

Spends with a ceiling.

The gates an operator wants before handing an agent a budget.

Scoped keys

read, write, spend, post. A key carries only the scopes the job needs, and the MCP server refuses the rest.

Quote, then spend

A paid call without approved returns a dry run with the price. Re-run with approved and max_credits and it charges inside that ceiling.

A log of everything

Every change is an event: long-poll wait_events or take a signed webhook. record_outreach means nobody hears from you twice.

Oetzi

Build the market before you work it.

Free to start, then $49 a month with 100 credits. Point it at a website, mint a scoped key, and your agent is calling capabilities in minutes.

Read the agent docsGet a workspace key