SidClaw

Approve, deny, and audit AI agent tool calls.

Works with MCP, LangChain, OpenAI Agents, Claude Agent SDK, and 15+ more.

Website · Documentation · Live Demo · SDK on npm · SDK on PyPI

---

Your agents call tools without oversight. SidClaw intercepts every tool call, checks it against your policies, and holds risky actions for human review before they execute.

See it in action

Agent wants to send an email → policy flags it → reviewer sees full context → approves or denies → trace recorded.

Works With Your Stack

SidClaw integrates with 18+ frameworks and platforms — including OpenClaw (329K+ users), LangChain, OpenAI, MCP, Claude Agent SDK, Google ADK, NemoClaw, Copilot Studio, GitHub Copilot, and more. Add governance in one line of code. See all integrations →

See It In Action

Customer Support Agent (Financial Services)

An AI agent wants to send a customer email. Policy flags it for review. The reviewer sees full context — who, what, why — and approves with one click. Every step is traced.

Infrastructure Automation (DevOps)

An AI agent wants to scale production services. High-risk deployments require human approval. Read-only monitoring is allowed instantly.

Clinical Decision Support (Healthcare)

An AI assistant recommends lab orders. The physician reviews the clinical context and approves. Medication prescribing is blocked by policy — only physicians can prescribe.

How It Works

Agent wants to act → SidClaw evaluates → Policy decides → Human approves (if needed) → Action executes → Trace recorded

Four primitives govern every agent action:

┌──────────┐    ┌──────────┐    ┌──────────┐    ┌──────────┐
│ Identity │ →  │  Policy  │ →  │ Approval │ →  │  Trace   │
│          │    │          │    │          │    │          │
│ Every    │    │ Every    │    │ High-risk│    │ Every    │
│ agent    │    │ action   │    │ actions  │    │ decision │
│ has an   │    │ evaluated│    │ get human│    │ creates  │
│ owner &  │    │ against  │    │ review   │    │ tamper-  │
│ scoped   │    │ explicit │    │ with rich│    │ proof    │
│ perms    │    │ rules    │    │ context  │    │ audit    │
└──────────┘    └──────────┘    └──────────┘    └──────────┘

• allow → action executes immediately, trace recorded
• approval_required → human sees context card, approves/denies, trace recorded
• deny → blocked before execution, no data accessed, trace recorded

Quick Start

npx create-sidclaw-app my-agent
cd my-agent
npm start

Add governance in one line

// Before: the agent decides, nobody reviews
await sendEmail({ to: "customer@example.com", subject: "Follow-up", body: "..." });

// After: wrap with SidClaw — now policies apply
const sendEmail = withGovernance(client, {
  operation: 'send_email',
  data_classification: 'confidential',
}, sendEmailFn);

await sendEmail({ to: "customer@example.com", subject: "Follow-up", body: "..." });
// → allow (executes) | approval_required (human reviews) | deny (blocked)

Same thing in Python

@with_governance(client, GovernanceConfig(
    operation="send_email",
    data_classification="confidential",
))
def send_email(to, subject, body):
    email_service.send(to=to, subject=subject, body=body)

Full TypeScript example with imports

npm install @sidclaw/sdk

import { AgentIdentityClient, withGovernance } from '@sidclaw/sdk';

const client = new AgentIdentityClient({
  apiKey: process.env.SIDCLAW_API_KEY,
  apiUrl: 'https://api.sidclaw.com',
  agentId: process.env.SIDCLAW_AGENT_ID,
});

const sendEmail = withGovernance(client, {
  operation: 'send_email',
  target_integration: 'email_service',
  resource_scope: 'customer_emails',
  data_classification: 'confidential',
}, async (to, subject, body) => {
  await emailService.send({ to, subject, body });
});

await sendEmail('customer@example.com', 'Follow-up', 'Hello...');
// allow → executes | approval_required → waits for human | deny → throws

Full Python example with imports

pip install sidclaw

import os
from sidclaw import SidClaw
from sidclaw.middleware.generic import with_governance, GovernanceConfig

client = SidClaw(
    api_key=os.environ["SIDCLAW_API_KEY"],
    agent_id=os.environ["SIDCLAW_AGENT_ID"],
)

@with_governance(client, GovernanceConfig(
    operation="send_email",
    target_integration="email_service",
    data_classification="confidential",
))
def send_email(to, subject, body):
    email_service.send(to=to, subject=subject, body=body)

MCP Governance Proxy

Wrap any MCP server with policy evaluation and approval workflows. Works with Claude Desktop, Cursor, VS Code, GitHub Copilot — any MCP client. Listed on the official MCP Registry.

Add to your .mcp.json:

{
  "mcpServers": {
    "postgres-governed": {
      "command": "npx",
      "args": ["-y", "@sidclaw/sdk", "sidclaw-mcp-proxy", "--transport", "stdio"],
      "env": {
        "SIDCLAW_API_KEY": "ai_your_key",
        "SIDCLAW_AGENT_ID": "your-agent-id",
        "SIDCLAW_UPSTREAM_CMD": "npx",
        "SIDCLAW_UPSTREAM_ARGS": "-y,@modelcontextprotocol/server-postgres,postgresql://localhost/mydb"
      }
    }
  }
}

• SELECT * FROM customers → allowed (~50ms overhead)
• DELETE FROM customers WHERE id = 5 → held for human approval
• DROP TABLE customers → denied by policy

Full MCP governance docs →

Why not just auth / sandboxing / logging?

Approach	What it solves	What it doesn't solve
Auth (Okta, OAuth)	Who is this agent?	Should this specific action execute right now?
Sandboxing (Docker, WASM)	Blast radius if something goes wrong	Whether the action should happen at all
Logging (Langfuse, LangSmith)	What happened after the fact	Intercepting actions before they execute
Policy engines (OPA)	General-purpose policy evaluation	Approval workflows, agent-specific context, audit trails
SidClaw	All of the above, plus the Approval primitive	—

SidClaw sits at the tool-call layer: the moment an agent decides to act in the real world.

Integrations

SidClaw wraps your existing agent tools — no changes to your agent logic.

Agent Frameworks

	TypeScript	Python
Core client	@sidclaw/sdk	sidclaw
MCP proxy	@sidclaw/sdk/mcp	sidclaw.mcp
LangChain	@sidclaw/sdk/langchain	sidclaw.middleware.langchain
OpenAI Agents	@sidclaw/sdk/openai-agents	sidclaw.middleware.openai_agents
CrewAI	@sidclaw/sdk/crewai	sidclaw.middleware.crewai
Vercel AI	@sidclaw/sdk/vercel-ai	—
Pydantic AI	—	sidclaw.middleware.pydantic_ai
Claude Agent SDK	@sidclaw/sdk/claude-agent-sdk	sidclaw.middleware.claude_agent_sdk
Google ADK	@sidclaw/sdk/google-adk	sidclaw.middleware.google_adk
LlamaIndex	@sidclaw/sdk/llamaindex	sidclaw.middleware.llamaindex
Composio	@sidclaw/sdk/composio	sidclaw.middleware.composio
NemoClaw	@sidclaw/sdk/nemoclaw	sidclaw.middleware.nemoclaw
Webhooks	@sidclaw/sdk/webhooks	sidclaw.webhooks

Platform Integrations

Integration	Description
Claude Code	Govern any MCP server in Claude Code. Add a .mcp.json entry — zero code changes. Guide →
OpenClaw	Governance proxy for OpenClaw skills. Published as sidclaw-governance on ClawHub. Guide →
MCP	Governance proxy for any MCP server. Listed on the official MCP Registry. CLI binary (sidclaw-mcp-proxy) + programmatic API. Guide →
NemoClaw	Govern NVIDIA NemoClaw sandbox tools with MCP-compatible proxy generation. Guide →
Copilot Studio	Governance for Microsoft Copilot Studio skills via OpenAPI action. Guide →
GitHub Copilot	Governance for GitHub Copilot agents via HTTP transport. Guide →
GitHub Action	sidclawhq/governance-action@v1 — reusable CI governance step. Guide →

Notification Channels

Approval requests are delivered to your team's preferred channels. Reviewers can approve or deny directly from chat.

Channel	Features
Slack	Block Kit messages with interactive Approve/Deny buttons. Messages update in-place after decision.
Microsoft Teams	Adaptive Card notifications with Approve/Deny buttons (Bot Framework) or dashboard links (webhook).
Telegram	HTML messages with inline keyboard. Callback updates remove buttons and add reply.
Resend	Email notifications for approval requests via transactional email.

Licensing

Component	License	What you can do
SDK (@sidclaw/sdk, sidclaw on PyPI)	Apache 2.0	Use freely, modify, distribute, commercial use
MCP Proxy (sidclaw-mcp-proxy)	Apache 2.0	Same as SDK
Platform (API, Dashboard, Docs)	FSL 1.1	Free for orgs under CHF 1M revenue. Converts to Apache 2.0 in 2028

Start with just the SDK? You don't need the platform. The SDK works standalone with the free hosted API at app.sidclaw.com, or you can self-host everything.

Why This Exists

AI agents are being deployed in production, but the governance layer is missing:

• 73% of CISOs fear AI agent risks, but only 30% are ready (NeuralTrust 2026)
• 79% of enterprises have blind spots where agents act without oversight
• FINRA 2026 explicitly requires "documented human checkpoints" for AI agent actions in financial services
• EU AI Act (August 2026) mandates human oversight, automatic logging, and risk management for high-risk AI systems
• OpenClaw has 329K+ stars and 13,700+ skills — but 1,184 malicious skills were found in the ClawHavoc campaign. There's no policy layer governing what skills can do.

The big vendors (Okta, SailPoint, WorkOS) handle identity and authorization. But none of them ship the approval step — the part where a human sees rich context and makes an informed decision before an agent acts.

Compliance

SidClaw maps to regulatory requirements across the US, EU, Switzerland, and Singapore:

🇺🇸 FINRA 2026 · 🇪🇺 EU AI Act · 🇨🇭 FINMA · 🇸🇬 MAS TRM · 🇺🇸 NIST AI RMF · 🌐 OWASP Agentic

Platform Features

For Developers

• 60-second setup — npx create-sidclaw-app scaffolds a working governed agent
• <50ms evaluation overhead — the governance layer is invisible to your users
• 5-minute integration — wrap existing tools, no code changes
• MCP-native — governance proxy for any MCP server
• Framework-agnostic — LangChain, Vercel AI, OpenAI, CrewAI, Pydantic AI, Composio, Claude Agent SDK, Google ADK, LlamaIndex, NemoClaw, or plain functions
• Typed SDKs — TypeScript (npm) + Python (PyPI)

For Security & Compliance Teams

• Policy engine — allow / approval_required / deny with priority ordering and classification hierarchy
• Approval workflow — context-rich cards with agent reasoning, risk classification, and separation of duties
• Audit trails — correlated traces with integrity hash chains (tamper-proof)
• SIEM export — JSON and CSV, continuous webhook delivery

For Platform Teams

• RBAC — admin, reviewer, viewer roles with enforced permissions
• Tenant isolation — automatic tenant scoping on every query
• API key management — scoped keys with rotation
• Rate limiting — per-tenant, per-endpoint-category
• Webhooks — real-time notifications for approvals, traces, lifecycle events
• Chat integrations — approve/deny from Slack, Teams, or Telegram without opening the dashboard
• Self-serve signup — GitHub, Google, email/password

Architecture

┌─────────────┐     ┌──────────────┐     ┌──────────────────┐
│ Your Agent  │     │  SidClaw SDK │     │  SidClaw API     │
│             │ ──► │              │ ──► │                  │
│ LangChain   │     │ evaluate()   │     │ Policy Engine    │
│ MCP Server  │     │ withGovern() │     │ Approval Service │
│ OpenAI SDK  │     │ governTools()│     │ Trace Store      │
│ Any tool    │     │              │     │ Webhook Delivery │
└─────────────┘     └──────────────┘     └──────────────────┘
                                                   │
                                          ┌────────┴────────┐
                                          ▼                 ▼
                                ┌──────────────┐  ┌──────────────┐
                                │  Dashboard   │  │ Notifications│
                                │              │  │              │
                                │ Agents       │  │ Slack        │
                                │ Policies     │  │ Teams        │
                                │ Approvals    │  │ Telegram     │
                                │ Traces       │  │ Email        │
                                │ Settings     │  │ Webhooks     │
                                └──────────────┘  └──────────────┘

Deploy

One-Click Deploy

Deploy from the GitHub repo to Railway. Add a PostgreSQL database, configure environment variables, and you're live.

Deploy the dashboard to Vercel (requires a separately hosted API).

Deploy Docs or Landing Page to Vercel

Docs:

Landing Page:

Self-Host (Docker)

curl -sSL https://raw.githubusercontent.com/sidclawhq/platform/main/deploy/self-host/setup.sh | bash

Or manually:

git clone https://github.com/sidclawhq/platform.git
cd platform
cp deployment/env.example .env  # edit with your values
docker compose -f docker-compose.production.yml up -d

Development credentials:

• Email: admin@example.com / Password: admin
• Or click "Sign in with SSO" on the login page to auto-login without a password

Hosted Cloud

No infrastructure to manage. Start free at app.sidclaw.com

See deployment documentation for production configuration, environment variables, and upgrade guides.

Documentation

• Quick Start — 2 minutes to first governed action
• SDK Reference — every method documented
• Integrations — MCP, OpenClaw, NemoClaw, LangChain, OpenAI, Claude Agent SDK, Google ADK, Copilot Studio, GitHub Copilot, and more
• Policy Guide — authoring, versioning, testing
• Compliance — 🇺🇸 FINRA · 🇪🇺 EU AI Act · 🇨🇭 FINMA · 🇸🇬 MAS TRM · 🇺🇸 NIST AI RMF · 🌐 OWASP
• API Reference — every endpoint

Contributing

We welcome contributions! See CONTRIBUTING.md for guidelines.

The SDK (packages/sdk/) is Apache 2.0. The platform (apps/) is FSL 1.1.

License

• SDK (packages/sdk/, packages/shared/): Apache License 2.0 — use freely for any purpose
• Platform (apps/api/, apps/dashboard/, apps/docs/, apps/landing/, apps/demo*/): Functional Source License 1.1 — source-available. Cannot offer as a competing hosted service. Converts to Apache 2.0 after 2 years (March 2028).

Links

• Website
• Documentation
• Dashboard
• TypeScript SDK (npm)
• Python SDK (PyPI)
• Python SDK (GitHub)
• create-sidclaw-app (npm)
• GitHub Action
• GitHub App
• Contact