Integration

Integration

Three ways to call the same 26 tools. All return the same PASS / FLAG / BLOCK + certificate envelope.

REST

No key; first 10 calls per IP are free.

curl -X POST https://api.agent-toolbox.ai/v1/scan/sql \
  -H "Content-Type: application/json" \
  -d '{"sql":"DELETE FROM users"}'

Discover pricing programmatically: GET https://api.agent-toolbox.ai/v1/pricing. Machine-readable spec: GET /openapi.json.

MCP

Adds all 26 tools to any MCP client (Claude Desktop, Cursor, Warp, VS Code):

{ "mcpServers": { "agent-toolbox": { "command": "npx", "args": ["-y", "agentoolbox-mcp"] } } }

No API key or env vars required — the server runs the tools in-process against free public data sources.

SDKs

TypeScript:

npm install agent-toolbox-sdk

import { AgentoolboxClient } from "agent-toolbox-sdk"; // types inferred from the OpenAPI contract
const client = new AgentoolboxClient();

Python — pip install agent-toolbox.ai (import module agent_toolbox).

The pattern: propose → validate → execute

Run the relevant gates before the action; act only on PASS (or FLAG, per policy). Example — a coding agent gating AI-generated code before install:

const secrets = await client.scanSecrets({ code });
if (!secrets.safe) throw new Error("hardcoded secrets");

const imports = await client.validateImports({ language: "python", code });
if (imports.hallucinated.length) throw new Error("hallucinated packages");
// …then scanVulnerabilities, verify, etc.

Independent gates can run in parallel with Promise.all. Store each certificate in your audit log. See the full list in the Tool Catalog.