Agent-native TypeScript framework for building MCP servers. Build tools, not infrastructure. Declarative definitions with auth, multi-backend storage, OpenTelemetry, and first-class support for Bun/Node/Cloudflare Workers.
@cyanheads/mcp-ts-core is the infrastructure layer for TypeScript MCP servers. Install it as a dependency — don't fork it. You write tools, resources, and prompts; the framework handles transports, auth, storage, config, logging, telemetry, and lifecycle.
import { createApp, tool, z } from '@cyanheads/mcp-ts-core';
const greet = tool('greet', {
description: 'Greet someone by name and return a personalized message.',
annotations: { readOnlyHint: true },
input: z.object({ name: z.string().describe('Name of the person to greet') }),
output: z.object({ message: z.string().describe('The greeting message') }),
handler: async (input) => ({ message: `Hello, ${input.name}!` }),
});
await createApp({ tools: [greet] });That's a complete MCP server. Every tool call is automatically logged with duration, payload sizes, memory usage, and request correlation — no instrumentation code needed. createApp() handles config parsing, logger init, transport startup, signal handlers, and graceful shutdown.
bunx @cyanheads/mcp-ts-core init my-mcp-server
cd my-mcp-server
bun installYou get a scaffolded project with CLAUDE.md, Agent Skills, and a src/ tree ready for your tools. Infrastructure — transports, auth, storage, telemetry, lifecycle, linting — lives in node_modules. What's left is domain: which APIs to wrap, which workflows to expose.
Start your coding agent (Claude Code, Codex, Cursor), describe the system you want to expose, and it drives the build. The included skills cover the full cycle: setup, design-mcp-server, scaffolding, testing, security-pass, release-and-publish.
Here's what tool definitions look like:
import { tool, z } from '@cyanheads/mcp-ts-core';
export const search = tool('search', {
description: 'Search for items by query.',
input: z.object({
query: z.string().describe('Search query'),
limit: z.number().default(10).describe('Max results'),
}),
output: z.object({ items: z.array(z.string()).describe('Search results') }),
async handler(input) {
const results = await doSearch(input.query, input.limit);
return { items: results };
},
});And resources:
import { resource, z } from '@cyanheads/mcp-ts-core';
export const itemData = resource('items://{itemId}', {
description: 'Retrieve item data by ID.',
params: z.object({ itemId: z.string().describe('Item ID') }),
async handler(params, ctx) {
return await getItem(params.itemId);
},
});Everything registers through createApp() in your entry point:
await createApp({
name: 'my-mcp-server',
version: '0.1.0',
tools: allToolDefinitions,
resources: allResourceDefinitions,
prompts: allPromptDefinitions,
});It also works on Cloudflare Workers with createWorkerHandler() — same definitions, different entry point.
- Declarative definitions —
tool(),resource(),prompt()builders with Zod schemas.appTool()/appResource()add interactive HTML UIs. - Unified Context — one
ctxfor logging, tenant-scoped storage, elicitation, sampling, cancellation, and task progress. - Inline auth —
auth: ['scope']on definitions. Framework checks scopes before dispatch — no wrapper code. - Task tools —
task: truefor long-running ops; framework manages create/poll/progress/complete/cancel. - Definition linter — validates names, schemas, auth scopes, annotation coherence, and format-parity at startup. Standalone CLI (
lint:mcp) and devcheck step. - Structured errors — handlers throw; framework catches, classifies, and formats. Error factories (
notFound(),validationError(), …) when the code matters. - Multi-backend storage —
in-memory, filesystem, Supabase, Cloudflare D1/KV/R2. Swap providers via env var; handlers don't change. - Pluggable auth —
none,jwt, oroauth. Local secret or JWKS verification. - Observability — Pino logging, optional OpenTelemetry traces and metrics. Request correlation and tool metrics are automatic.
- Local + edge — same definitions run on stdio, HTTP (Hono), and Cloudflare Workers.
- Tiered dependencies — parsers, OTEL SDK, Supabase, and OpenAI are optional peers. Install what you use.
- Agent-first DX — ships
CLAUDE.mdwith the full exports catalog so AI agents ramp up without prompting.
my-mcp-server/
src/
index.ts # createApp() entry point
worker.ts # createWorkerHandler() (optional)
config/
server-config.ts # Server-specific env vars
services/
[domain]/ # Domain services (init/accessor pattern)
mcp-server/
tools/definitions/ # Tool definitions (.tool.ts)
resources/definitions/ # Resource definitions (.resource.ts)
prompts/definitions/ # Prompt definitions (.prompt.ts)
package.json
tsconfig.json # extends @cyanheads/mcp-ts-core/tsconfig.base.json
CLAUDE.md # Points to core's CLAUDE.md for framework docs
No src/utils/, no src/storage/, no src/types-global/, no src/mcp-server/transports/ — infrastructure lives in node_modules.
All core config is Zod-validated from environment variables. Server-specific config uses a separate Zod schema with lazy parsing.
| Variable | Description | Default |
|---|---|---|
MCP_TRANSPORT_TYPE |
stdio or http |
stdio |
MCP_HTTP_PORT |
HTTP server port | 3010 |
MCP_HTTP_HOST |
HTTP server hostname | 127.0.0.1 |
MCP_AUTH_MODE |
none, jwt, or oauth |
none |
MCP_AUTH_SECRET_KEY |
JWT signing secret (required for jwt mode) |
— |
STORAGE_PROVIDER_TYPE |
in-memory, filesystem, supabase, cloudflare-d1/kv/r2 |
in-memory |
OTEL_ENABLED |
Enable OpenTelemetry | false |
OPENROUTER_API_KEY |
OpenRouter LLM API key | — |
See CLAUDE.md for the full configuration reference.
| Function | Purpose |
|---|---|
createApp(options) |
Node.js server — handles full lifecycle |
createWorkerHandler(options) |
Cloudflare Workers — returns { fetch, scheduled } |
| Builder | Usage |
|---|---|
tool(name, options) |
Define a tool with handler(input, ctx) |
resource(uriTemplate, options) |
Define a resource with handler(params, ctx) |
prompt(name, options) |
Define a prompt with generate(args) |
appTool(name, options) |
Define an MCP Apps tool with auto-populated _meta.ui |
appResource(uriTemplate, options) |
Define an MCP Apps HTML resource with the correct MIME type and _meta.ui mirroring for read content |
Handlers receive a unified Context object:
| Property | Type | Description |
|---|---|---|
ctx.log |
ContextLogger |
Request-scoped logger (auto-correlates requestId, traceId, tenantId) |
ctx.state |
ContextState |
Tenant-scoped key-value storage |
ctx.elicit |
Function? |
Ask the user for input (when client supports it) |
ctx.sample |
Function? |
Request LLM completion from the client |
ctx.signal |
AbortSignal |
Cancellation signal |
ctx.notifyResourceUpdated |
Function? |
Notify subscribed clients a resource changed |
ctx.notifyResourceListChanged |
Function? |
Notify clients the resource list changed |
ctx.progress |
ContextProgress? |
Task progress reporting (when task: true) |
ctx.requestId |
string |
Unique request ID |
ctx.tenantId |
string? |
Tenant ID (from JWT or 'default' for stdio) |
import { createApp, tool, resource, prompt } from '@cyanheads/mcp-ts-core';
import { createWorkerHandler } from '@cyanheads/mcp-ts-core/worker';
import { McpError, JsonRpcErrorCode, notFound, serviceUnavailable } from '@cyanheads/mcp-ts-core/errors';
import { checkScopes } from '@cyanheads/mcp-ts-core/auth';
import { markdown, fetchWithTimeout } from '@cyanheads/mcp-ts-core/utils';
import { OpenRouterProvider, GraphService } from '@cyanheads/mcp-ts-core/services';
import { validateDefinitions } from '@cyanheads/mcp-ts-core/linter';
import { createMockContext } from '@cyanheads/mcp-ts-core/testing';
import { fuzzTool, fuzzResource, fuzzPrompt } from '@cyanheads/mcp-ts-core/testing/fuzz';See CLAUDE.md for the complete exports reference.
The examples/ directory contains a reference server consuming core through public exports, demonstrating all patterns:
| Tool | Pattern |
|---|---|
template_echo_message |
Basic tool with format, auth |
template_cat_fact |
External API call, error factories |
template_madlibs_elicitation |
ctx.elicit for interactive input |
template_code_review_sampling |
ctx.sample for LLM completion |
template_image_test |
Image content blocks |
template_async_countdown |
task: true with ctx.progress |
template_data_explorer |
MCP Apps with linked UI resource via appTool()/appResource() builders |
import { createMockContext } from '@cyanheads/mcp-ts-core/testing';
import { myTool } from '@/mcp-server/tools/definitions/my-tool.tool.js';
const ctx = createMockContext({ tenantId: 'test-tenant' });
const input = myTool.input.parse({ query: 'test' });
const result = await myTool.handler(input, ctx);createMockContext() provides stubbed log, state, and signal. Pass { tenantId } for state operations, { sample } for LLM mocking, { elicit } for elicitation mocking, { progress: true } for task tools.
Schema-aware fuzz testing via fast-check. Generates valid inputs from Zod schemas and adversarial payloads (prototype pollution, injection strings, type confusion) to verify handler invariants.
import { fuzzTool } from '@cyanheads/mcp-ts-core/testing/fuzz';
const report = await fuzzTool(myTool, { numRuns: 100 });
expect(report.crashes).toHaveLength(0);
expect(report.leaks).toHaveLength(0);
expect(report.prototypePollution).toBe(false);Also exports fuzzResource, fuzzPrompt, zodToArbitrary, and ADVERSARIAL_STRINGS for custom property-based tests.
- CLAUDE.md — Framework reference: exports catalog, patterns, Context interface, error codes, auth, config, testing. Ships in the npm package.
- CHANGELOG.md — Version history
bun run rebuild # clean + build (scripts/clean.ts + scripts/build.ts)
bun run devcheck # lint, format, typecheck, MCP defs, audit, outdated
bun run lint:mcp # validate MCP definitions against spec
bun run test:all # vitest (unit + integration)Issues and pull requests welcome. Run checks before submitting:
bun run devcheck
bun run test:allApache 2.0 — see LICENSE.