gRPC in Multimeter, the AI-powered REST Client for VS Code.
Markdown (no JavaScript)
# gRPC ```yaml type: api protocol: grpc url: grpc://localhost:50051 grpc: service: helloworld.Greeter method: SayHello message: name: "World" outputs: greeting: message.message ``` Notes: - `protocol: grpc` is required — use `grpc://` or `grpcs://` URL schemes - The `grpc` block replaces `body`, `method`, `format`, `query`, and `cookies` - `grpc.proto` defaults to `reflect` (server reflection); set it to a `.proto` file path for file-based definitions - `grpc.stream` supports `server`, `client`, or `bidi`; omit for unary calls - `headers` are sent as gRPC metadata - Outputs use `message.*` and `metadata.*` extraction roots - `auth` maps to metadata: bearer → `authorization: Bearer <token>`, basic → `authorization: Basic <encoded>` ### grpc block | Field | Type | Required | Description | |-----------|--------|----------|-------------| | `proto` | string | No | Path to `.proto` file, or `"reflect"` (default) | | `service` | string | Yes | Fully-qualified service name (e.g. `helloworld.Greeter`) | | `method` | string | Yes | RPC method name (e.g. `SayHello`) | | `stream` | string | No | `server`, `client`, or `bidi`. Omit for unary | | `message` | object | No | Request message as key-value pairs | URL: `grpc://host:port` (plaintext) or `grpcs://host:port` (TLS). ```yaml grpc: proto: ./protos/greeter.proto service: helloworld.Greeter method: SayHello outputs: greeting: message.message requestId: metadata.x-request-id ``` `body.*` / `headers.*` also work and map to `message` / `metadata`. ### Streaming ```yaml grpc: service: chat.ChatService method: StreamMessages stream: server message: channel: general ``` For server streaming, `message` contains all collected messages as an array. Client and bidi streaming send the message object as a single frame. See also: [HTTP](./http.md) · [Outputs](../outputs.md) · [Auth](../auth.md)
For AI agents · llms.txt · Home