Web pages for your agent, with hidden text removed.
One call returns any public page as clean Markdown, strips what people cannot see and flags text aimed at the model. 500 free calls, no account, no card.
One call
curl -X POST https://angel.oroboros-design.com/stealth-read -H 'content-type: application/json' -H 'X-Bundle-Token: <your key>' -d '{"url":"https://example.com"}'
The answer is JSON: markdown (headings, lists and tables kept; navigation, cookie banners, scripts and everything hidden from people removed), title, lang, verdict and findings for text aimed at the model, and removed with what was taken out.
MCP
{"mcpServers":{"angel":{"url":"https://angel.oroboros-design.com/agenten/signal/mcp","headers":{"X-Bundle-Token":"<your key>"}}}}
Tool stealth_read. Without a key, 5 calls a day work over MCP and A2A.
Prices
The same key works on stealth-read, input-guard, preflight, token-risk and shield-check. Details: pricing.
Limits
- Angel reads as a declared bot, without login and without bypassing bot protection. About one page in five blocks it. A blocked page is not counted.
- The checks are rule-based. They find hidden content well and attacks written in plain language weakly. A clean result is one signal, not proof.
For agents
llms.txt · OpenAPI · Agent card (A2A) · How hidden text reaches a model