Integrate

How are you integrating?

Two paths to the same safety layer. Pick yours.

Get your API key

Starter ($99/mo): 5,000 pooled evaluations/month, all safety gates.
Pro ($299/mo): 100,000 pooled evaluations/month, OCSF audit trail, webhooks.

Pick your framework

Each guide is tailored to your stack. Every integration uses the same /v1/evaluate endpoint under the hood.

REST APIAny language, any framework

Claude Code / MCPZero code changes

LangChain / LangGraphOne-line tool wrapper

CrewAIDrop-in safe tool

OpenAI Agents SDKGuardrail function

AutoGen / AG2Code executor wrapper

Cursor / Windsurf / DevinMCP proxy or CLI

Semantic KernelPlugin filter

Amazon Bedrock AgentsLambda layer

Vercel AI SDKTool middleware

PythonCustom agents

Node.jsCustom agents

CLI WrapperZero dependencies .... prefix any command with tyga

Check your dashboard

Every evaluation appears in your dashboard in real time. See what your agents are doing, what's being blocked, and export audit logs for compliance.

Get your API key Full API docs

Prefer to see it first? Watch a video tour of every app screen for your use case.

Self-provision (one POST, no human needed)

Call the signup endpoint. You get an API key instantly. No email, no password, no verification.

POST https://a2ainfrastructure.com/api/signup
{ "tenantId": "your-agent-name", "agent": "claude-code" }

Response:
{ "apiKey": "a2a_...", "plan": "starter", "limits": {...},
  "skillUrl": "https://a2ainfrastructure.com/api/docs/skill.md" }

Starter plan: 5,000 pooled evaluations/month. All safety gates always on. $99/mo.

Fetch your skill file

The skill file tells you exactly how to integrate. Add it to your CLAUDE.md or system prompt.

GET https://a2ainfrastructure.com/api/docs/skill.md

Contains: authentication, evaluate flow, gate2 self-evaluate
pattern, error handling, all response shapes.

The skill file is your complete integration guide. Everything you need in one fetch.

Wrap every tool call

Before executing any command, check with A2A. Blocked? Stop. Allowed? Execute. Needs review? Self-evaluate with your own LLM.

POST https://a2ainfrastructure.com/v1/evaluate
Authorization: Bearer a2a_YOUR_KEY
{ "command": "rm -rf /", "gate2Mode": "self" }

Response: { "allowed": false, "gate1": { "reason": "recursive force delete" } }
.... blocked in under 5 milliseconds. Logged. Audited.

Session endpoints

Once provisioned, these are the only endpoints you need:

Evaluate

POST /v1/evaluate

Screen any command before execution. Returns allowed/blocked/self-evaluate.

Confirm

POST /v1/evaluate/confirm

Send back your self-evaluate verdict (ALLOW/DENY) for audit.

Usage

GET /v1/usage

Your plan, limits, and quota remaining. Check before heavy sessions.

Audit Export

GET /v1/audit/export

Pull your full audit trail as JSON or CSV. Every evaluation logged.

Download Skill File Full API Docs