Developers

Wrap a tool. Sentinel does the rest.

Agents call Sentinel before they act. The SDKs handle tokens, delegation, approvals and tracing, and fail closed. Or put the MCP proxy in front of your MCP servers and change no agent code.

Get developer access Developer sign-in

TypeScript

A guarded tool in a few lines

The wrapped function runs only after Sentinel allows the call, or after a person approves it. In co-pilot mode it receives the arguments the approver chose.

TypeScriptPythonGoJava 
const sentinel = new SentinelClient({
  baseUrl: 'https://sentinel.example.com',
  agentId: 'agt_01HZX…',
  clientSecret: process.env.AGENT_SECRET,
  principal: 'user:1234'
});

// Runs only after Sentinel allows it, or a person approves.
const refund = sentinel.wrapTool('payments', 'refund',
  issueRefund, { waitForApproval: true });

await refund({ amount: 750, currency: 'USD' });

Ways to integrate

Pick the path that fits your stack.

SDKs

TypeScript, Python, Go and Java clients: token refresh, delegation to sub-agents, authorization, approval waits and W3C trace context.

Framework adapters

Ready adapters for the Claude Agent SDK, OpenAI Agents, LangChain and LlamaIndex: each tool call goes through Sentinel.

MCP proxy

Point agents at Sentinel's MCP endpoint instead of your servers. Every tool call is authorized, and tool results can be checked by content guardrails.

Workload identity

Agents on Kubernetes or AWS exchange their platform identity for a Sentinel token, with no stored secret.

Events and webhooks

Every state change is an event; webhooks are signed with HMAC-SHA256 and can post to Slack, Teams or your SIEM.

Integration contract

A versioned contract for authorization, signals, approvals and revocations, with JSON Schemas and a public JWKS for token verification.

How a call flows

Five steps, a few milliseconds

  1. The agent gets a short-lived token (or a delegated one from its parent).
  2. Before a tool call, it asks Sentinel to authorize the call with its arguments.
  3. Sentinel checks every layer and answers allow, deny, or wait for a person.
  4. The decision is written to the audit chain before the tool runs.
  5. Signals from content guardrails, and revocations, reach the agent's session in seconds.

Developer access

Docs and a sandbox

Approved developers get the full documentation (quickstart, API and token reference, policies, the integration contract, and the TypeScript, Python, Go and Java SDK guides) and a private sandbox tenant with a 30-day API token.

Request developer access · Sign in