REAL RUN · 2026-07-22

Claude summarized a public Burn demo Vault.

Six distinct English-titled articles. One read-only MCP call. Claude grouped the saved sources into three themes and chose what to read next—through the production burn-mcp-server.

Download the iPhone app Reproduce this run Inspect the evidence

01 / EVIDENCE

Inspect the captured call, not a mockup.

The replay renders the captured terminal run. The call and result are verbatim.

Real run, publish-safe replay. English neural narration and captions included; waiting time is compressed and the hidden token was revoked.

EXACT PROMPT
You are recording a public Burn MCP demo. Respond in English only. Call the Burn list_vault tool exactly once with category AI Tools and limit 11. Do not call any other tool. From the returned metadata, select exactly 6 distinct items whose titles contain no Chinese, Japanese, or Korean characters, removing duplicate titles. Based only on those six items, write exactly three concise themes. Then recommend exactly three of those saved articles to read next, using their exact English titles and one short reason each. Do not quote, translate, or mention excluded or duplicate titles. Finish with this exact sentence: Evidence: 11 Vault items retrieved; 6 distinct English-titled items examined.
OBSERVED MCP CALLS
  1. 01list_vault({"category":"AI Tools","limit":11})

VERIFIED CLAUDE RESULT

Retrieved 11 Vault items from the AI Tools category. After removing one duplicate title and the items with non-English titles, exactly six distinct English-titled items remain, and everything below is based only on those six.

Monetizing AI agents directly — several pieces frame OpenClaw not as a tool but as a revenue engine, from $70k/month B2C app growth to a $100K/month productized service playbook.

Fully automated content machines — recurring focus on hands-off pipelines that generate, repurpose, and distribute content across many platforms and accounts.

Agent setup and reliability fundamentals — before scaling, the groundwork matters: persistent memory, rule enforcement, and token-cost control keep agents from breaking or burning cash.

Before You Do Anything With OpenClaw, Read This (You're Welcome) — the foundation piece: it fixes agent amnesia, rule erosion, and runaway token costs, which every other workflow depends on.

How to turn OpenClaw into a $100K/month business — the most concrete monetization playbook, with specific pricing for setup services and recurring content pipelines.

Turn your OpenClaw into a Content Machine (send this to your agent) — a complete human-in-the-loop pipeline that turns one podcast episode into six-plus assets, with the tool integrations spelled out.

Evidence: 11 Vault items retrieved; 6 distinct English-titled items examined.

6 DISTINCT ENGLISH-TITLED ITEMS EXAMINED

  1. how to ACTUALLY build an ai automated cashflow content machine...
  2. How our OpenClaw agent "Eddie" helps us make $70k/mo with B2C Apps ( step by step )
  3. Before You Do Anything With OpenClaw, Read This (You're Welcome)
  4. Turn your OpenClaw into a Content Machine (send this to your agent)
  5. How to turn OpenClaw into a $100K/month business
  6. how to make money so fast with OpenClaw it feels ILLEGAL 🦞💰

02 / CONNECT

Reproduce it in about three minutes.

The recorded path uses the npm package over local stdio—the standard Claude Desktop setup.

  1. 01

    Generate a Burn MCP token

    In Burn for iPhone, open Settings → MCP Server → Generate token. The web token page is also available to password-based accounts. Never publish or record the token.

    Open the web token page Download the iPhone app
  2. 02

    Add burn-mcp-server to Claude

    Paste this into Claude Desktop's MCP configuration. Node.js is required.

    {
  "mcpServers": {
    "burn": {
      "command": "npx",
      "args": ["burn-mcp-server"],
      "env": {
        "BURN_MCP_TOKEN": "<your-token>"
      }
    }
  }
}
  3. 03

    Use the same read-only prompt

    You are recording a public Burn MCP demo. Respond in English only. Call the Burn list_vault tool exactly once with category AI Tools and limit 11. Do not call any other tool. From the returned metadata, select exactly 6 distinct items whose titles contain no Chinese, Japanese, or Korean characters, removing duplicate titles. Based only on those six items, write exactly three concise themes. Then recommend exactly three of those saved articles to read next, using their exact English titles and one short reason each. Do not quote, translate, or mention excluded or duplicate titles. Finish with this exact sentence: Evidence: 11 Vault items retrieved; 6 distinct English-titled items examined.

Does your client support Streamable HTTP?

Use the hosted endpoint with an Authorization header instead of running Node locally.

{
  "mcpServers": {
    "burn": {
      "type": "http",
      "url": "https://burn-mcp-server.vercel.app/api/mcp",
      "headers": {
        "Authorization": "Bearer <your-token>"
      }
    }
  }
}

Hosted endpoint: https://burn-mcp-server.vercel.app/api/mcp

03 / LIMITS

What this demo proves—and what it does not.

Trust starts with a narrow claim that survives inspection.

Existing category, real server
This recording retrieves 11 public article metadata records from the existing Vault category AI Tools, then excludes four non-English titles and one duplicate to examine 6 distinct English-titled items. Replace AI Tools with a category from your own Vault; your results will differ and stay private unless you send them to your AI client.
Metadata, not article bodies
list_vault returns titles, URLs, tags, categories, positioning, and takeaways. This run did not request extracted article text.
Keyword search is separate
search_vault matches titles, tags, and takeaways. It is keyword search, not semantic or vector retrieval.
Write tools exist
Triage, collection, watched-source, and analysis tools can write. This demo allowed only list_vault and called no write tool.