# Custom adapters

Create a custom adapter when the prebuilt server adapters (Hono, Express, Fastify, Koa) don't support your framework or you have specific request/response handling requirements.

A custom adapter translates between Mastra's route definitions and your framework's routing system. You'll implement methods that register middleware, handle requests, and send responses using your framework's APIs.

> **Info:** Use any of these prebuilt server adapters:
>
> - [@mastra/hono](https://mastra.ai/reference/server/hono-adapter)
> - [@mastra/express](https://mastra.ai/reference/server/express-adapter)
> - [@mastra/fastify](https://mastra.ai/reference/server/fastify-adapter)
> - [@mastra/koa](https://mastra.ai/reference/server/koa-adapter)

## Abstract class

The `MastraServer` abstract class from `@mastra/server/server-adapter` provides the foundation for all adapters. It handles route registration logic, parameter validation, and other shared functionality. Your custom adapter extends this class and implements the framework-specific parts.

The class takes three type parameters that represent your framework's types:

```typescript
import { MastraServer } from '@mastra/server/server-adapter'

export class MyFrameworkServer extends MastraServer<
  // Your framework's app type (e.g., FastifyInstance)
  MyApp,
  // Your framework's request type (e.g., FastifyRequest)
  MyRequest,
  // Your framework's response type (e.g., FastifyReply)
  MyResponse
> {
  // Implement abstract methods
}
```

These type parameters ensure type safety throughout your adapter implementation and enable proper typing when accessing framework-specific APIs.

## Required methods

You must implement these six abstract methods. Each handles a specific part of the request lifecycle, from attaching context to sending responses.

### `registerContextMiddleware()`

This method runs first and attaches Mastra context to every incoming request. Route handlers need access to the Mastra instance, tools, and other context to function. How you attach this context depends on your framework — Express uses `res.locals`, Hono uses `c.set()`, and other frameworks have their own patterns.

```typescript
registerContextMiddleware(): void {
  this.app.use('*', (req, res, next) => {
    // Attach context to your framework's request/response
    res.locals.mastra = this.mastra;
    res.locals.requestContext = new RequestContext();
    res.locals.tools = this.tools;
    res.locals.abortSignal = createAbortSignal(req);
    next();
  });
}
```

Context to attach:

| Key              | Type                   | Description                      |
| ---------------- | ---------------------- | -------------------------------- |
| `mastra`         | `Mastra`               | The Mastra instance              |
| `requestContext` | `RequestContext`       | Request-scoped context map       |
| `tools`          | `Record<string, Tool>` | Available tools                  |
| `abortSignal`    | `AbortSignal`          | Request cancellation signal      |
| `taskStore`      | `InMemoryTaskStore`    | A2A task storage (if configured) |

### `registerAuthMiddleware()`

Register authentication and authorization middleware. This method should check if authentication is configured on the Mastra instance and skip registration entirely if not. When auth is configured, you'll typically register two middleware functions: one for authentication (validating tokens and setting the user) and one for authorization (checking if the user can access the requested resource).

```typescript
registerAuthMiddleware(): void {
  const authConfig = this.mastra.getServer()?.auth;
  if (!authConfig) return;

  // Register authentication (validate token, set user)
  this.app.use('*', async (req, res, next) => {
    const token = extractToken(req);
    const user = await authConfig.authenticateToken?.(token, req);
    if (!user) {
      return res.status(401).json({ error: 'Unauthorized' });
    }
    res.locals.user = user;
    next();
  });

  // Register authorization (check permissions)
  this.app.use('*', async (req, res, next) => {
    const allowed = await authConfig.authorize?.(
      req.path,
      req.method,
      res.locals.user,
      res
    );
    if (!allowed) {
      return res.status(403).json({ error: 'Forbidden' });
    }
    next();
  });
}
```

### `registerRoute()`

Register a single route with your framework. This method is called once for each Mastra route during initialization. It receives a `ServerRoute` object containing the path, HTTP method, handler function, and Zod schemas for validation. Your implementation should wire this up to your framework's routing system.

```typescript
async registerRoute(
  app: MyApp,
  route: ServerRoute,
  { prefix }: { prefix?: string }
): Promise<void> {
  const path = `${prefix || ''}${route.path}`;
  const method = route.method.toLowerCase();

  app[method](path, async (req, res) => {
    try {
      // 1. Extract parameters
      const params = await this.getParams(route, req);

      // 2. Validate with Zod schemas
      const queryParams = await this.parseQueryParams(route, params.queryParams);
      const body = await this.parseBody(route, params.body);

      // 3. Build handler params
      const handlerParams = {
        ...params.urlParams,
        ...queryParams,
        ...(typeof body === 'object' ? body : {}),
        mastra: this.mastra,
        requestContext: res.locals.requestContext,
        tools: res.locals.tools,
        abortSignal: res.locals.abortSignal,
        taskStore: this.taskStore,
      };

      // 4. Call handler
      const result = await route.handler(handlerParams);

      // 5. Send response
      return this.sendResponse(route, res, result);
    } catch (error) {
      const status = error.status ?? error.details?.status ?? 500;
      return res.status(status).json({ error: error.message });
    }
  });
}
```

### `getParams()`

Extract URL parameters, query parameters, and request body from the incoming request. Different frameworks expose these values in different ways—Express uses `req.params`, `req.query`, and `req.body`, while other frameworks may use different property names or require method calls. This method normalizes the extraction for your framework.

```typescript
async getParams(
  route: ServerRoute,
  request: MyRequest
): Promise<{
  urlParams: Record<string, string>;
  queryParams: Record<string, string>;
  body: unknown;
}> {
  return {
    // From route path (e.g., :agentId)
    urlParams: request.params,
    // From URL query string
    queryParams: request.query,
    // From request body
    body: request.body,
  };
}
```

### `sendResponse()`

Send the response back to the client based on the route's response type. Mastra routes can return different response types: JSON for most API responses, streams for agent generation, and special types for MCP transports. Your implementation should handle each type appropriately for your framework.

```typescript
async sendResponse(
  route: ServerRoute,
  response: MyResponse,
  result: unknown
): Promise<unknown> {
  switch (route.responseType) {
    case 'json':
      return response.json(result);

    case 'stream':
      return this.stream(route, response, result);

    case 'datastream-response':
      // Return AI SDK Response directly
      return result;

    case 'mcp-http':
      // Handle MCP HTTP transport
      return this.handleMcpHttp(response, result);

    case 'mcp-sse':
      // Handle MCP SSE transport
      return this.handleMcpSse(response, result);

    default:
      return response.json(result);
  }
}
```

### `stream()`

Handle streaming responses for agent generation. When an agent generates a response, it produces a stream of chunks that should be sent to the client as they become available. This method reads from the stream, optionally applies redaction to hide sensitive data, and writes chunks to the response in the appropriate format (SSE or newline-delimited JSON).

```typescript
async stream(
  route: ServerRoute,
  response: MyResponse,
  result: unknown
): Promise<unknown> {
  const isSSE = route.streamFormat === 'sse';

  // Set streaming headers based on format
  response.setHeader('Content-Type', isSSE ? 'text/event-stream' : 'text/plain');
  response.setHeader('Transfer-Encoding', 'chunked');

  const reader = result.fullStream.getReader();

  try {
    while (true) {
      const { done, value } = await reader.read();
      if (done) break;

      // Apply redaction if enabled
      const chunk = this.streamOptions.redact
        ? redactChunk(value)
        : value;

      // Format based on stream format
      if (isSSE) {
        response.write(`data: ${JSON.stringify(chunk)}\n\n`);
      } else {
        response.write(JSON.stringify(chunk) + '\x1E');
      }
    }

    // Send completion marker (SSE uses data: [DONE], other formats use record separator)
    if (isSSE) {
      response.write('data: [DONE]\n\n');
    }
    response.end();
  } catch (error) {
    reader.cancel();
    throw error;
  }
}
```

## Helper methods

The base class provides helper methods you can use in your implementation. These handle common tasks like parameter validation and route registration, so you don't need to reimplement them:

| Method                                                              | Description                                                 |
| ------------------------------------------------------------------- | ----------------------------------------------------------- |
| `parsePathParams(route, params)`                                    | Validate path params with Zod schema                        |
| `parseQueryParams(route, params)`                                   | Validate query params with Zod schema                       |
| `parseBody(route, body)`                                            | Validate body with Zod schema                               |
| `mergeRequestContext({ paramsRequestContext, bodyRequestContext })` | Merge request context from multiple sources                 |
| `registerRoutes()`                                                  | Register all Mastra routes (calls `registerRoute` for each) |
| `registerOpenAPIRoute(app, config, { prefix })`                     | Register OpenAPI spec endpoint                              |

The `parse*` methods use Zod schemas defined on each route to validate input and return typed results. If validation fails, they throw an error with details about what went wrong.

## Constructor

Your adapter's constructor should accept the same options as the base class and pass them to `super()`. You can add additional framework-specific options if needed:

```typescript
constructor(options: {
  app: MyApp;
  mastra: Mastra;
  prefix?: string;
  openapiPath?: string;
  bodyLimitOptions?: BodyLimitOptions;
  streamOptions?: StreamOptions;
  customRouteAuthConfig?: Map<string, boolean>;
}) {
  super(options);
}
```

See [Server Adapters](https://mastra.ai/docs/server/server-adapters) for full documentation on each option.

## Full example

Here's a skeleton implementation showing all the required methods. This uses pseudocode for framework-specific parts—replace with your framework's actual APIs:

```typescript
import { MastraServer, ServerRoute } from '@mastra/server/server-adapter'
import type { Mastra } from '@mastra/core'

export class MyFrameworkServer extends MastraServer<MyApp, MyRequest, MyResponse> {
  constructor(options: { app: MyApp; mastra: Mastra; prefix?: string }) {
    super(options)
  }

  registerContextMiddleware(): void {
    this.app.use('*', (req, res, next) => {
      res.locals.mastra = this.mastra
      res.locals.requestContext = this.mergeRequestContext({
        paramsRequestContext: req.query.requestContext,
        bodyRequestContext: req.body?.requestContext,
      })
      res.locals.tools = this.tools ?? {}
      res.locals.abortSignal = createAbortSignal(req)
      next()
    })
  }

  registerAuthMiddleware(): void {
    const authConfig = this.mastra.getServer()?.auth
    if (!authConfig) return
    // ... implement auth middleware
  }

  async registerRoute(
    app: MyApp,
    route: ServerRoute,
    { prefix }: { prefix?: string },
  ): Promise<void> {
    // ... implement route registration
  }

  async getParams(route: ServerRoute, request: MyRequest) {
    return {
      urlParams: request.params,
      queryParams: request.query,
      body: request.body,
    }
  }

  async sendResponse(route: ServerRoute, response: MyResponse, result: unknown) {
    if (route.responseType === 'stream') {
      return this.stream(route, response, result)
    }
    return response.json(result)
  }

  async stream(route: ServerRoute, response: MyResponse, result: unknown) {
    // ... implement streaming
  }
}
```

## Usage

Once your adapter is implemented, use it the same way as the provided adapters:

```typescript
import { MyFrameworkServer } from './my-framework-adapter'
import { mastra } from './mastra'

const app = createMyFrameworkApp()
const server = new MyFrameworkServer({ app, mastra })

await server.init()

app.listen(4111)
```

> **Tip:** The existing [@mastra/hono](https://github.com/mastra-ai/mastra/blob/main/server-adapters/hono/src/index.ts) and [@mastra/express](https://github.com/mastra-ai/mastra/blob/main/server-adapters/express/src/index.ts) implementations are good references when building your custom adapter. They show how to handle framework-specific patterns for context storage, middleware registration, and response handling.
>
> If you want to use [Studio](https://mastra.ai/docs/studio/overview) with your server adapter, use [`mastra studio`](https://mastra.ai/reference/cli/mastra) to only launch the Studio UI.

## Related

- [Server Adapters](https://mastra.ai/docs/server/server-adapters): Overview and shared concepts
- [Hono Adapter](https://mastra.ai/reference/server/hono-adapter): Reference implementation
- [Express Adapter](https://mastra.ai/reference/server/express-adapter): Reference implementation
- [MastraServer Reference](https://mastra.ai/reference/server/mastra-server): Full API reference
- [createRoute() Reference](https://mastra.ai/reference/server/create-route): Creating type-safe custom routes