Comprehensive guide for Effect-TS, the functional TypeScript library. Use when building Effect applications, especially MCP servers. Covers correct APIs, common misconceptions, and MCP-specific patterns.
Effect-TS is a functional TypeScript library providing typed effects, structured concurrency, and a robust runtime. This skill covers correct usage patterns and addresses common misconceptions from LLM-generated content.
import { Effect, Layer, Context, Fiber, Schedule, Cache, Scope } from 'effect';
import { Schema, JSONSchema } from '@effect/schema';
Core Type Signature:
Effect<Success, Error, Requirements>;
// ↑ ↑ ↑
// | | └── Dependencies (provided via Layers)
// | └── Expected errors (typed, must be handled)
// └── Success value
LLM outputs often contain incorrect APIs. Use this table to correct them:
| Wrong (common in AI outputs) | Correct |
|---|---|
Effect.cachedWithTTL(...) | Cache.make({ timeToLive: Duration }) |
Effect.cachedInvalidateWithTTL(...) | cache.invalidate(key) / cache.invalidateAll() |
Effect.match(...) | Effect.either + Either.match, or Effect.catchTag |
| "thread-local storage" | "fiber-local storage" via FiberRef |
| JSON Schema Draft 2020-12 | @effect/schema generates Draft-07 |
| fibers are "cancelled" | fibers are "terminated" or "interrupted" |
| all queues have back-pressure | only bounded queues; sliding/dropping do not |
--only=production | --omit=dev (npm 7+) |
Effect distinguishes between:
E channel, must be handled// Expected error - typed
const fetchUser = (id: string): Effect.Effect<User, UserNotFoundError | NetworkError> => ...
// Handle expected errors
const handled = pipe(
fetchUser("123"),
Effect.catchTag("UserNotFoundError", (e) => Effect.succeed(defaultUser)),
Effect.catchTag("NetworkError", (e) => Effect.retry(schedule))
);
// Unexpected errors (defects) - captured by runtime
Effect.catchAllDefect(program, (defect) =>
Console.error("Unexpected error", defect)
);
Fibers are lightweight virtual threads with native interruption:
// Fork a fiber
const fiber = yield * Effect.fork(longRunningTask);
// Interrupt it (e.g., when MCP client disconnects)
yield * Fiber.interrupt(fiber);
// Structured concurrency: child fibers auto-terminate with parent
const parent = Effect.gen(function* () {
yield* Effect.fork(backgroundTask); // Auto-interrupted when parent ends
yield* mainTask;
});
// Daemon fibers outlive their parent
yield * Effect.forkDaemon(longLivedBackgroundTask);
// First to succeed wins; other is automatically interrupted
const result = yield * Effect.race(fetchFromCache, fetchFromDatabase);
// Process 50 documents with max 5 concurrent
const results =
yield *
Effect.all(documents.map(processDoc), {
concurrency: 5, // NOT a "worker pool" - limits concurrent tasks
});
// Bounded - applies back-pressure (offer suspends when full)
const bounded = yield * Queue.bounded<string>(100);
// Dropping - discards new items when full (no back-pressure)
const dropping = yield * Queue.dropping<string>(100);
// Sliding - discards oldest items when full (no back-pressure)
const sliding = yield * Queue.sliding<string>(100);
Layers construct services without leaking dependencies:
// Define a service
class Database extends Context.Tag('Database')<
Database,
{ query: (sql: string) => Effect.Effect<Result> }
>() {}
// Create layer (dependencies handled at construction)
const DatabaseLive = Layer.effect(
Database,
Effect.gen(function* () {
const config = yield* Config; // Dependency injected here
return {
query: sql => Effect.tryPromise(() => runQuery(sql, config)),
};
}),
);
// Provide to program
const runnable = program.pipe(Effect.provide(DatabaseLive));
// For testing - swap implementation
const DatabaseTest = Layer.succeed(Database, {
query: () => Effect.succeed(mockResult),
});
const program = pipe(
Effect.tryPromise(() => openConnection()),
Effect.ensuring(Console.log('Cleanup')), // Runs on success, failure, OR interrupt
);
const withConnection = Effect.acquireUseRelease(
Effect.tryPromise(() => db.connect()), // Acquire
conn => Effect.tryPromise(() => conn.query('SELECT *')), // Use
conn => Effect.promise(() => conn.close()), // Release (always runs)
);
Effect.scoped(
Effect.gen(function* () {
const file = yield* openFile('data.txt'); // Acquired
const data = yield* file.read();
return data;
}), // File automatically released when scope closes
);
There is no Effect.cachedWithTTL. Use the Cache module:
import { Cache } from 'effect';
const cache =
yield *
Cache.make({
capacity: 100,
timeToLive: Duration.minutes(5),
lookup: (key: string) => fetchExpensiveData(key),
});
// Use the cache
const value = yield * cache.get('my-key');
// Invalidate
yield * cache.invalidate('my-key');
yield * cache.invalidateAll();
import { Schedule } from 'effect';
// Retry 3 times with exponential backoff
const policy = Schedule.exponential('100 millis').pipe(
Schedule.intersect(Schedule.recurs(3)),
);
const robust = Effect.retry(unstableOperation, policy);
// Retry until condition
const untilSuccess = Effect.retry(operation, {
until: err => err.code === 'RATE_LIMITED',
});
@effect/schema generates JSON Schema Draft-07 (not 2020-12):
import { Schema, JSONSchema } from '@effect/schema';
const User = Schema.Struct({
id: Schema.String,
age: Schema.Number.pipe(Schema.positive()),
});
// Generate JSON Schema (Draft-07)
const jsonSchema = JSONSchema.make(User);
// { "$schema": "http://json-schema.org/draft-07/schema#", ... }
// Decode (parse)
const user = Schema.decodeUnknownSync(User)(rawData);
// Encode
const json = Schema.encodeSync(User)(user);
Effect has native OpenTelemetry integration:
import { NodeSdk } from '@effect/opentelemetry';
import { OTLPTraceExporter } from '@opentelemetry/exporter-trace-otlp-http';
// Add tracing to any effect
const traced = Effect.withSpan('processRequest')(myEffect);
// Logging with context
yield * Effect.log('Processing request');
yield * Effect.annotateLogs('requestId', 'abc-123');
// FiberRef for fiber-local context propagation
const RequestId = FiberRef.unsafeMake<string>('');
yield * FiberRef.set(RequestId, 'req-456');
| Scenario | Recommendation |
|---|---|
| Simple MCP tool (< 100 LOC) | Use FastMCP or vanilla SDK |
| Team unfamiliar with FP | Steep learning curve; consider NestJS |
| Bundle size critical | Effect adds 15-25kb gzipped minimum |
| Existing NestJS/TypeORM codebase | Impedance mismatch with class-based DI |
const searchTool = Effect.gen(function* () {
const args = yield* parseArgs(input);
const db = yield* Database;
const results = yield* db.query(args.query);
return formatResults(results);
}).pipe(
Effect.catchTag('ParseError', () =>
Effect.fail({ code: -32602, message: 'Invalid params' }),
),
Effect.catchTag('DatabaseError', () =>
Effect.fail({ code: -32603, message: 'Internal error' }),
),
);
// Each MCP request gets its own scope
const handleRequest = (request: MCPRequest) =>
Effect.scoped(
Effect.gen(function* () {
// Resources acquired here auto-release when request completes
const tempFile = yield* createTempFile();
const result = yield* processRequest(request, tempFile);
return result;
}),
);