Cosmo connect

Cosmo Connect

Cosmo Connect lets you use GraphQL Federation without running GraphQL servers. Define an Apollo-compatible Subgraph Schema, compile it to protobuf, implement it as gRPC.

Decision: Plugin vs gRPC Service

Factor	Router Plugin	gRPC Service
Languages	Go, TypeScript (Bun)	Any gRPC language
Deployment	Co-located with router	Independent microservice
Scaling	Coupled to router	Independent
Latency	Minimal (IPC)	Network overhead
Team autonomy	Low	High

Use plugins for simple integrations, lowest latency, wrapping legacy APIs. Use gRPC services for team ownership, independent scaling, language flexibility.

Core Workflow

1. Define GraphQL Schema (standard Subgraph Schema)
2. Generate Protobuf + Mapping (wgc CLI)
3. Implement gRPC logic
4. Configure Router
5. Compose supergraph (wgc router compose)
6. Update README with example queries (see "Example Queries in README" section)
7. Deploy and test

Plugin Workflow

# Scaffold (Go)
wgc router plugin init my-plugin -p myproject --language go

# Or scaffold (TypeScript)
wgc router plugin init my-plugin -p myproject --language ts

# Edit schema: my-plugin/src/schema.graphql
# Generate proto + code:
wgc router plugin generate ./my-plugin

# Implement resolvers in src/main.go or src/plugin.ts
# Test:
wgc router plugin test ./my-plugin

# Build:
wgc router plugin build ./my-plugin

# Publish to Cosmo Cloud:
wgc router plugin publish ./my-plugin

gRPC Service Workflow

# Scaffold
wgc grpc-service init --template golang-connect-rpc --directory ./my-service
# Or: --template typescript-connect-rpc-fastify

# Edit schema, then generate:
wgc grpc-service generate -i ./schema.graphql -o ./generated

# Implement service, then publish:
wgc grpc-service publish my-service --schema ./schema.graphql --generated ./generated

Go Plugin Implementation Pattern

package main

import (
    "log"
    service "github.com/wundergraph/cosmo/plugin/generated"
    routerplugin "github.com/wundergraph/cosmo/router-plugin"
    "google.golang.org/grpc"
)

func main() {
    pl, err := routerplugin.NewRouterPlugin(func(s *grpc.Server) {
        s.RegisterService(&service.MyService_ServiceDesc, &MyServiceImpl{})
    })
    if err != nil {
        log.Fatalf("failed to create router plugin: %v", err)
    }
    pl.Serve()
}

TypeScript Plugin Implementation Pattern

import * as grpc from '@grpc/grpc-js';
import { MyServiceService, IMyServiceServer } from '../generated/service_grpc_pb.js';
import { PluginServer } from './plugin-server.js';

class MyService implements IMyServiceServer {
    [name: string]: grpc.UntypedHandleCall;
    queryData(call, callback) {
        const response = new QueryResponse();
        callback(null, response);
    }
}

const pluginServer = new PluginServer();
pluginServer.addService(MyServiceService, new MyService());
pluginServer.serve();

Go gRPC Service Pattern (Connect RPC)

type Service struct{}

func (s *Service) QueryGetProject(ctx context.Context, req *connect.Request[pb.QueryGetProjectRequest]) (*connect.Response[pb.QueryGetProjectResponse], error) {
    return connect.NewResponse(&pb.QueryGetProjectResponse{
        GetProject: &pb.Project{Id: req.Msg.Id, Name: "My Project"},
    }), nil
}

TypeScript gRPC Service Pattern (Connect RPC + Fastify)

import type { ConnectRouter } from "@connectrpc/connect";
import { Service } from "./proto/service/v1/service_pb.js";

export default (router: ConnectRouter) => {
    router.service(Service, {
        queryGetProject: async (req) => ({
            getProject: { id: req.id, name: "My Project" },
        }),
    });
};

Field Resolvers

For fields with arguments, use @connect__fieldResolver:

type Foo {
    id: ID!
    bar(baz: String!): String! @connect__fieldResolver(context: "id")
}

Generates a dedicated ResolveFooBar RPC. Batching is automatic. Critical: Return results in the exact same order as provided context elements.

Local Router Setup (No Cloud)

Set ROUTER_CONFIG_PATH to run in static config mode — no controlplane or API token needed.

# 1. Create graph.yaml declaring subgraphs (HTTP, gRPC standalone, or plugin)
# 2. Compose into execution config:
wgc router compose -i ./graph.yaml -o ./config.json

# 3. Run router:
ROUTER_CONFIG_PATH=./config.json go run ./cmd/router/main.go

# For plugins, also set:
PLUGINS_ENABLED=true PLUGINS_PATH=./plugins

graph.yaml Subgraph Shapes

# Standard GraphQL (HTTP)
- name: my-subgraph
  routing_url: http://localhost:4001/graphql
  schema:
    file: ./path/to/schema.graphqls

# Connect - Standalone gRPC
- name: my-grpc-subgraph
  routing_url: dns:///localhost:4011
  grpc:
    schema_file: ./path/to/src/schema.graphql
    proto_file: ./path/to/generated/service.proto
    mapping_file: ./path/to/generated/mapping.json

# Connect - Plugin
- name: my-plugin-subgraph
  plugin:
    version: 0.0.1
    path: ./path/to/plugin-dir

Router Config (config.yaml)

version: "1"
plugins:
  enabled: true
  path: "plugins"

GraphQL Feature Support

Supported: Query, Mutation, Entity Lookups (single/multiple/compound keys), Scalar/Complex args, Enums, Interfaces, Unions, Recursive types, Nested objects, Lists, Entity definitions, Key directives, External fields, Field resolvers (with batching).

Not yet supported: Nested key entity lookups, @requires, Subscriptions, Nullable list items (protobuf constraint).

Example Queries in README

After composing the supergraph schema (wgc router compose), always add example queries to the project's README. This lets users immediately copy-paste queries into the playground.

When to generate

After every wgc router compose that succeeds, check the README for an "Example Queries" section and update it to reflect the current composed schema.

How to generate

1. Read the composed supergraph config (config.json) or the individual subgraph schemas to identify all root Query and Mutation fields.
2. For each root field, write a minimal but useful example query that includes:
   • The operation name (PascalCase matching the field name)
   • All required arguments with variables
   • A representative selection of return fields (not every field — pick the most useful 3-5)
   • For federation entity joins, include the cross-subgraph fields with a comment noting which subgraph resolves them
3. Add a variables JSON block for any query/mutation that requires input arguments.
4. Group queries before mutations. Lead with the simplest query.

README format

## Example Queries

<description of the query>:

\`\`\`graphql
query OperationName {
  fieldName {
    field1
    field2
    field3
  }
}
\`\`\`

<description including federation details if applicable>:

\`\`\`graphql
query OperationName($input: InputType!) {
  fieldName(arg: $input) {
    field1
    field2
    crossSubgraphField  # resolved from <other-subgraph> plugin
  }
}
\`\`\`

Variables:
\`\`\`json
{
  "input": { ... }
}
\`\`\`

More examples are available in the `examples/` directory.

Rules

• Only include queries/mutations that exist in the composed schema — don't guess.
• Keep each example minimal — show the shape, not every field.
• For union return types, show both ... on SuccessType and ... on ErrorType fragments.
• If an examples/ directory exists with .graphql files, reference it at the end.
• Don't duplicate the full examples directory content — pick 2-3 representative queries for the README.

References

• For CLI command details, see references/cli-reference.md
• For Go development details (plugins + HTTP client + logging + tracing), see references/go-development.md
• For TypeScript development details, see references/ts-development.md
• For local development setup (router, graph.yaml, standalone/plugin step-by-step, troubleshooting), see references/local-development.md
• For internal ConnectRPC architecture, see references/architecture.md