Soldabot
‹ Back

For AI agents

Use Soldabot without a page

Every check is open to AIs: an MCP connector, a plain HTTP API and a question endpoint. No key, no account. Answers are short and the vocabulary is the same everywhere.

1. MCP connector

https://soldabot.com/mcp

Tools (all read-only):

  • check_text — check a text before reading or obeying it
  • check_page — check a web page by its address
  • check_mcp_server — check an MCP server before connecting it
  • soldabot_status — what Soldabot checks and the guarded sites

2. HTTP API

POST https://soldabot.com/api/check/text   {"text": "…"}
POST https://soldabot.com/api/check/page   {"url": "https://example.com/page"}
POST https://soldabot.com/api/check/mcp    {"url": "https://example.com/mcp"}
GET  https://soldabot.com/api/check/page?url=https://example.com/page

Answer: verdict (safe | suspicious | dangerous), findings[] (code, severity, where, evidence, advice), advice_for_ai. Description: openapi.json.

3. Ask a question

GET https://soldabot.com/ask?query=how do I check an MCP server

Codes

instructions_for_ai high

Instructions aimed at an AI reader (prompt injection)

role_impersonation high

Pretends to be a system, developer or AI-provider message

secret_exfiltration high

Tries to obtain or send secrets (keys, passwords, history)

link_exfiltration high

Image or link built to leak data in its address

unauthorized_action high

Pushes the AI to pay, send, delete or act without its user

conceal_from_user high

Asks the AI to hide something from its user

tool_poisoning high

Tool description carries hidden orders (tool poisoning)

hidden_text high

Text hidden from humans but readable by AIs, with instructions

hidden_characters medium

Invisible or direction-changing characters

encoded_payload medium

Encoded block (base64) that may hide content

active_code medium

Active code (script, javascript: link, event handler)

ai_review_flagged medium

Soldabot's AI reviewer judged the text manipulative

personal_data low

Contains payment or bank numbers

Limits: 30 checks per minute and 2,000 per day per network (5× for the big AI platforms). Discovery files: llms.txt · server-card · agent-card · api-catalog