Handling Authentication with Your API

Many APIs require authentication (e.g., API keys, Bearer tokens) to access their resources. AgentB allows you to securely connect your agents to these protected APIs.

Goal: Modify the AgentB server to handle an API that requires a Bearer token, and demonstrate how the AgentB facade can manage request-specific authentication.

Prerequisites

Completed Tutorial 2: Connecting AgentB to Your API.
You have an API endpoint that requires a Bearer token (or you can simulate one). For this tutorial, we'll conceptually use a "secure" version of a service.

The Challenge: Dynamic Authentication

Often, the authentication token (like a user-specific Bearer token) isn't static. It might come from:

A user's session.
An HTTP Authorization header in the request made to your AgentB server.
A secure vault or configuration specific to the user making the request.

AgentB's AgentB.getExpressStreamingHttpHandler (and its core logic) supports a powerful authorizeRequest callback that can return PerProviderAuthOverrides. This allows your AgentB server to dynamically provide authentication details for specific tool providers on a per-request basis.

Step 1: Modify Your OpenAPI Specification (Conceptual)

Let's assume you have an OpenAPI specification for a "My Secure Service" that requires a Bearer token.

specs/my-secure-service.json (Conceptual Excerpt):

{
  "openapi": "3.0.0",
  "info": {
    "title": "My Secure Service",
    "version": "1.0.0"
  },
  "servers": [{ "url": "https://api.mysecure.service/v1" }], // Replace with actual URL
  "components": {
    "securitySchemes": {
      "BearerAuth": { // Define the security scheme
        "type": "http",
        "scheme": "bearer",
        "bearerFormat": "JWT" // Optional, but good practice
      }
    }
  },
  "security": [ // Apply the security scheme globally or per-operation
    {
      "BearerAuth": []
    }
  ],
  "paths": {
    "/data": {
      "get": {
        "summary": "Get some secure data",
        "operationId": "getSecureData",
        "responses": {
          "200": { "description": "Secure data retrieved" }
          // ...
        }
        // If this operation needed specific scopes, they'd be listed under "BearerAuth" here.
      }
    }
  }
}

The key parts are components.securitySchemes defining BearerAuth and the top-level security applying it.

Step 2: Update Your AgentB Server

We'll modify the server.ts from previous tutorials.

import * // ... your existing imports ...
import { AgentB, ToolProviderSourceConfig, LLMMessage, PerProviderAuthOverrides } from '@ulifeai/agentb'; // Add PerProviderAuthOverrides

// ... (Express app setup, CORS, etc. as before) ...

async function startServer() {
  // ... (AgentB.initialize as before) ...
  AgentB.initialize({
    llmProvider: { provider: 'openai', model: 'gpt-4o-mini' },
  });

  // 1. Register your "My Secure Service"
  let mySecureServiceSpec;
  const fs = require('fs'); // Make sure to require fs and path
  const path = require('path');
  try {
    // Assume specs/my-secure-service.json exists
    const specPath = path.join(__dirname, '../specs/my-secure-service.json'); // Adjust path
    const specFileContent = fs.readFileSync(specPath, 'utf-8');
    mySecureServiceSpec = JSON.parse(specFileContent);

    const secureApiProviderConfig: ToolProviderSourceConfig = {
      id: 'mySecureService', // This ID is crucial for auth overrides
      type: 'openapi',
      openapiConnectorOptions: {
        spec: mySecureServiceSpec,
        sourceId: 'mySecureService', // Match the 'id' above
        // Static authentication can be a fallback or placeholder
        // We will override this dynamically per request.
        authentication: {
          type: 'bearer',
          token: '' // Static token is empty, expecting override
        }
      },
      toolsetCreationStrategy: 'allInOne',
      allInOneToolsetName: 'SecureServiceTools'
    };
    AgentB.registerToolProvider(secureApiProviderConfig);
    console.log("🛠️ Registered My Secure Service Tool Provider.");

  } catch (error) {
    console.error("❌ Failed to load or register My Secure Service spec:", error);
    console.log("Ensure 'specs/my-secure-service.json' exists and is valid for this tutorial.");
  }

  // 2. Update the HTTP Handler with `authorizeRequest`
  app.post('/agent/stream', AgentB.getExpressStreamingHttpHandler({
    getThreadId: async (req: Request, threadStorage) => { /* ... as before ... */
      const requestedThreadId = req.body.threadId || req.query.threadId as string;
      if (requestedThreadId) {
        const threadExists = await threadStorage.getThread(requestedThreadId);
        if (threadExists) return requestedThreadId;
      }
      const newThread = await threadStorage.createThread({ title: `Web Chat ${new Date().toISOString()}` });
      return newThread.id;
    },
    getUserMessage: async (req: Request) => req.body.prompt,

    // ⭐ New: authorizeRequest callback
    authorizeRequest: async (req: Request, threadId: string): Promise<boolean | PerProviderAuthOverrides> => {
      console.log(`[Auth] Authorizing request for thread: ${threadId}, path: ${req.path}`);

      const overrides: PerProviderAuthOverrides = {};
      let isAuthorized = false; // Start with not authorized

      // Example: Extract Bearer token from Authorization header
      const authHeader = req.headers.authorization;
      if (authHeader && authHeader.startsWith('Bearer ')) {
        const token = authHeader.substring(7); // Remove "Bearer " prefix
        console.log(`[Auth] Found Bearer token in request for 'mySecureService'.`);

        // Provide this token specifically for the 'mySecureService' provider
        overrides['mySecureService'] = {
          type: 'bearer',
          token: token,
        };
        isAuthorized = true; // Request is authorized if token is present
      } else {
        console.warn("[Auth] No Bearer token found in Authorization header for 'mySecureService'. This provider might fail if it requires auth.");
        // Depending on your app's logic:
        // - You might still return `true` if the request is generally allowed but the tool might fail.
        // - Or, if the token is absolutely required for any useful interaction with this provider,
        //   you could return `false` here to deny the request outright.
        // For this example, we'll allow the request to proceed, and the tool call will likely fail if auth is truly needed by the API.
        // If NO auth mechanism is found for a provider that needs it, but other providers can be used,
        // you might still return true or the overrides object.
        // If *this specific request requires* the secure service and auth is missing, then return false.
        // For simplicity, if the prompt indicates secure data is needed and no token, we could deny.
        if (req.body.prompt?.toLowerCase().includes("secure data") && !isAuthorized) {
            console.log("[Auth] Prompt asks for secure data but no token. Denying.");
            return false; // Deny the request
        }
        isAuthorized = true; // Allow general chat even if secure token is missing for some tools
      }

      // Add logic for other providers if needed:
      // const apiKey = req.headers['x-another-api-key'] as string;
      // if (apiKey) {
      //   overrides['anotherProviderId'] = { type: 'apiKey', name: 'X-API-KEY', in: 'header', key: apiKey };
      //   isAuthorized = true; // If any auth is successful
      // }

      if (!isAuthorized && Object.keys(overrides).length === 0) {
        // If no auth mechanisms were successful for any provider needing auth,
        // and the request is deemed unauthorized overall.
        // This logic depends on your application's requirements.
        // For this example, we made `isAuthorized = true` if *any* auth for *any* provider was found,
        // or even if no auth was found but it's not strictly required for *this request*.
      }
      
      // If the request is authorized to proceed:
      // - Return `true` if no dynamic overrides are needed (static config will be used).
      // - Return the `overrides` object if dynamic auth details were determined.
      // - Return `false` to deny the request (e.g., HTTP 403 Forbidden).
      return Object.keys(overrides).length > 0 ? overrides : isAuthorized;
    },
  }));

  // ... (Error handler and app.listen as before) ...
  app.listen(PORT, () => {
    console.log(`✅ AgentB Express server (with Auth) listening on port ${PORT}`);
    // ...
  });
}

startServer().catch(console.error);

Step 3: Test the Authenticated Endpoint

Restart your AgentB server (node server.js or npx ts-node server.ts).
Use curl or Postman to interact.

Scenario 1: No Authentication Token

curl -X POST \
  -H "Content-Type: application/json" \
  -d '{"prompt":"Can you fetch my secure data?"}' \
  http://localhost:3001/agent/stream \
  --no-buffer

Expected Server Logs (Conceptual):

[Auth] Authorizing request for thread: ..., path: /agent/stream
[Auth] No Bearer token found in Authorization header for 'mySecureService'. ...
[Auth] Prompt asks for secure data but no token. Denying.

(The client would receive an HTTP error, likely 403 Forbidden, or whatever your AgentB.getExpressStreamingHttpHandler sends when authorizeRequest returns false).If authorizeRequest returned true instead of false in this case, the flow would continue, and the LLM might try to use the tool. The actual API call by OpenAPIConnector would then fail if the (empty) static token isn't valid for "My Secure Service".

Scenario 2: With a Valid (or placeholder) Authentication Token

curl -X POST \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_ACTUAL_OR_PLACEHOLDER_TOKEN" \
  -d '{"prompt":"Please get my secure data using the mySecureService tools."}' \
  http://localhost:3001/agent/stream \
  --no-buffer

Replace YOUR_ACTUAL_OR_PLACEHOLDER_TOKEN with a token.

Expected Server Logs (Conceptual):

[Auth] Authorizing request for thread: ..., path: /agent/stream
[Auth] Found Bearer token in request for 'mySecureService'.
... (AgentB event stream starts) ...
[LLM Intends to Call Tool: mySecureService_getSecureData with args: {}]
[Tool Executing: mySecureService_getSecureData with input {}...]
... (OpenAPIConnector will use the provided token for the API call) ...
[Tool Result (mySecureService_getSecureData) -> Success: true] (if API call succeeds)

Expected Client Output (curl): You'll see the usual AgentB event stream. If the API call made by the mySecureService_getSecureData tool is successful (because the token was valid for the target API), the agent will receive the data and formulate a response. If the token is invalid for the actual "My Secure Service" API, the tool execution will fail, and the agent will report that.

Key Takeaways

authorizeRequest Callback: This function in AgentB.getExpressStreamingHttpHandler(options) is your central point for request-level authorization and dynamic authentication.
PerProviderAuthOverrides: The authorizeRequest callback can return this object. The keys are the ids you defined in your ToolProviderSourceConfig (e.g., 'mySecureService'). The values are ConnectorAuthentication objects ({ type: 'bearer', token: '...' }, { type: 'apiKey', ... }, etc.).
Dynamic Token Injection: OpenAPIConnector (used internally by AgentB for OpenAPI tools) will pick up these dynamic overrides for the specific API call, overriding any static authentication configured for that provider.
Security: Your AgentB server acts as a secure gateway. Client applications send requests to your server, which then securely attaches the necessary credentials before the agent's tools call the downstream APIs. Never expose raw user tokens directly to the LLM or store them insecurely.

This pattern is powerful for multi-tenant applications or scenarios where each user request to your AgentB server might need to use different credentials for the underlying tools/APIs.

Next Up: Creating a Custom Tool that doesn't rely on an OpenAPI specification.

PreviousAdding the Chat UI (@ulifeai/agentb-ui)NextCreating a Custom Tool

Last updated 29 days ago