Power your agent
पब्लिशर की अपनी कमांड। PlugMCP एंडपॉइंट नहीं।
Host a copy with our keys and logs.
npx -y [email protected]Claude Code में लोकल कमांड के लिए इंस्टॉल लिंक नहीं है। नीचे दी गई कमांड कॉपी करें और टर्मिनल में चलाएँ।
claude mcp add orpc-fn -- npx -y orpc-fn@0.9.0टर्मिनल में चलाएँ।
{ "mcpServers": { "orpc-fn": { "command": "npx", "args": [ "-y", "[email protected]" ] } }}claude_desktop_config.json
रिमोट URL या आपकी मशीन पर कमांड? दोनों इंस्टॉल में क्या अंतर है
PlugMCP पर चलाएँ
A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.
लॉन्चपब्लिशर से
Define a backend function once. You get a typed oRPC procedure with tracing, structured logging, guards, handler extras from your app, nested calls that keep the context, and optional live queries over Redis. Adapters expose the same procedure as a Mastra tool, an MCP tool, or a route on Hono, and the client helpers call it from Vue, React or Expo.
Define a backend function once: a typed oRPC procedure with tracing, logging, guards, injected handler extras, nested calls and live queries over Redis.
Your app keeps everything app-specific: auth and session, context shape, database, queues, i18n, feature flags and permissions. You inject them once with createFn.
bun add orpc-fn @orpc/server @orpc/contract @orpc/client zod
ESM only, Node 20+ or Bun, TypeScript 5.7+, oRPC 1.14+ (@orpc/client is already a dependency of @orpc/server), Zod 4. Optional peers per subpath are listed below.
// lib/fn.ts — once per app
import { ORPCError, os } from '@orpc/server'
import * as otel from '@opentelemetry/api'
import { createFn } from 'orpc-fn'
const publicProcedure = os.$context<{ headers: Headers; user?: User }>()
const protectedProcedure = publicProcedure.use(({ context, next }) => {
if (!context.user) throw new ORPCError('UNAUTHORIZED')
return next({ context: { user: context.user } })
})
export const { fn, fnLive, createPubSub, createRouter } = createFn({
procedures: { public: publicProcedure, protected: protectedProcedure },
default: 'protected',
extras: () => ({ db }),
guards: {
permission: (required: Permission, { context }) => context.user?.can(required) ?? false
},
otel
})
// features/orders/get-order.ts
export const getOrder = fn({
name: 'order.get',
method: 'GET',
path: '/orders/{id}',
permission: 'orders:read',
input: z.object({ id: z.string() }),
handler: async ({ input, context, db, call, logger }) => {
logger.info('loading order', { id: input.id }) // context.user is typed: protected
const order = await db.order.find(input.id)
const customer = await call(getCustomer, { id: order.customerId }) // child span, same context
return { ...order, customer }
}
})
Every call runs in an OpenTelemetry span named order.get (for a streaming handler, until the stream ends), runs the permission guard, observes input/output validation, and logs one fn.completed line at info, warn (4xx) or error (defects).
orpc-fn: createFncreateFn({
procedures, // named oRPC builders; the handler context is the builder's context
default, // builder used when a route omits `procedure`; leave out to require `procedure` on every route
extras, // ({ context, span, signal, name, procedure }) => values (or a promise) merged into handler params
extrasByProcedure, // e.g. { protected: ({ context }) => ({ t: translator(context.user) }) }
guards, // { key: (value, params) => void | boolean | Promise<void | boolean> } — each key is a typed fn() option
meta, // defineMeta<AppMeta>() — allowed values of each route's meta object
tags, // ['internal', 'external'] — allowed route tags
otel, // `import * as otel from '@opentelemetry/api'`; omit for no tracing
logger, // (scope, span) => your logger (debug/info/warn/error)
spanAttributes, // ({ context, name, procedure, meta }) => extra span attributes
onCompleted, // (event) => extra attributes for the fn.completed log
isExpectedError, // default: 4xx ORPCError or AbortError
pubsub // { transport (or () => transport, created on first use), namespace?, ownsTransport?, maxQueueSize?, maxIngressSize?, maxReplaySize?, initializationTimeoutMs?, onDrop?, onMetric? } for fnLive / createPubSub
})
It returns fn, fnLive, createPubSub, createPublisher, createRouter, readMeta, createCall, shutdown, drainPubSubSubscribers and activePubSubSubscriberCount.
Procedure builders must not set an input or output schema (os.input(...)); put schemas on routes. Guard keys cannot reuse route option names. Application metadata lives in a typed meta object and may share names with guards or route options. Unknown top-level route options and extras that shadow built-in handler params are rejected. Scoped extras replace matching shared keys, with handler types reflecting that precedence.
fn(options) has four overloads (input and output schema, each given or inferred). Route options: name (required; operationId and span name), procedure, path, method, summary, description, tags, deprecated, successStatus, successDescription, inputStructure, outputStructure, configured guard keys, guardResolvers, meta, stream and native errors declarations. Core schemas use Standard Schema; Zod 4 remains the default for tool/live adapters. The handler receives input, context, call, signal, span, logger, native typed errors, lastEventId and your extras. With an output schema, the handler returns the schema's input type and callers receive its output type, so output: z.string().transform(Number) means the handler returns a string and callers get a number.
Guards run inside the span, before the handler, and only on routes that set their option. Return false or throw an ORPCError to refuse; true and undefined allow. Static guards run sequentially in route declaration order, followed by resolved guards in resolver order. Unsupported return values fail types and runtime. For input-dependent checks use guardResolvers: { permission: ({ input, context }) => requirement }; these see parsed input. A guard cannot set both a static value and a resolver.
call(procedure, input) calls another procedure with the caller's context and abort signal, inside a call: <name> child span. Input and output are typed from the callee. Calling a procedure whose context the caller's context does not satisfy is a type error. Outside a handler (seeders, scripts, tests), use createCall(context, signal?) from createFn.
Meta is stored in oRPC's own procedure meta. It survives os.router(), prefixes and lazy routers. Declare the contract with defineMeta<AppMeta>() (no runtime defaults), then use fn({ meta: { readOnly: true }, ... }). Both readMeta(procedure) and readFnMeta(procedure) infer metadata from the passed procedure, including procedures from another factory.
Completion hooks discriminate by procedure and handlerStarted. Before handler context exists (auth/input failure), they receive the initial context. Logging, tracing, metrics and completion-hook failures preserve procedure results. Custom log fields cannot overwrite canonical completion fields. Stream completion runs once when the stream ends.
Also exported: defineMeta, createStreamManifest, createStreamManifestAsync, renderStreamManifest, isExpectedClientError, createBoundedEventQueue, createRouter, createBoundCall, and the types HandlerParams, RouteConfig, BoundCall, FnMeta, FnContext, FnCompletedEvent, ProcedureInput, ProcedureOutput, ProcedureContext, InitialContextOf, CurrentContextOf and GuardOptions.
orpc-fn/live: pub/sub and live queriesimport { createFn } from 'orpc-fn'
import { bunRedisTransport } from 'orpc-fn/live/redis-bun' // or ioredis / memory
export const { fnLive, createPubSub, drainPubSubSubscribers } = createFn({
/* … */,
pubsub: {
// A function defers the connection until the first publish or subscribe.
transport: () => bunRedisTransport(new RedisClient(process.env.REDIS_URL)),
ownsTransport: true,
onDrop: (count) => dropCounter.add(count)
}
})
export const listOrders = fnLive({
name: 'order.list',
input: z.object({ organizationId: z.string() }),
handler: ({ input, db }) => db.order.list(input.organizationId),
live: {
eventSchema: z.object({ organizationId: z.string(), orderId: z.string() }),
channel: ({ organizationId }) => `orders:${organizationId}`,
coalesceMs: 50
}
})
// router: { list: listOrders.procedure, subscribe: listOrders.subscribe }
await listOrders.publish({ organizationId, orderId }) // after a write
The subscribe route runs builder middleware and the same fn guards/extras/meta/completion policy, then authFn and subscribes to the channel, then streams the initial snapshot, then reloads when invalidated (once per coalescing batch). Queue bounds (maxQueueSize, maxIngressSize, maxReplaySize) can be overridden per pub/sub definition or live config. maxBatchSize bounds a live coalescing batch (default 1000). pubsub.onMetric reports typed queue depths, drops, successful reconnects and parsing failures; keep this synchronous callback fast. Startup waits for a successful broker subscription, bounded by initializationTimeoutMs (10000ms by default) and cancellation. A refused subscriber receives nothing, and events published while the snapshot loads are not lost. Snapshots go through the output schema, like the route's own result. Options:
shouldUpdate skips an event.transformerFn folds an event into initialized previous. It requires a stateSchema that validates parsed state. Patches returned by fnLivePatch(state, emit) additionally require emitSchema. These schemas should not transform already-parsed values. rerun() rechecks the current input through the procedure; it cannot change subscriber input.coalesceMs performs one rerun for an invalidation batch. Reducers process every event but emit their final full state; ordered patch payloads remain ordered. Batches are bounded (1000 events by default in streamLiveSnapshots).authFn and useBacklog/backlogSize/backlogTtl work as on createPubSub.SERVICE_UNAVAILABLE after detected loss, or emit an application-provided overflowMarker that must trigger recovery.reauthorize can check access before each live update. Initial authorization covers idle streams; apps that need immediate revocation should close those streams.safePublish logs publish errors instead of throwing them; it does not retry or guarantee delivery.createPubSub returns a typed subscribe route plus publish, publishMany (one atomic round-trip across channels) and typed getSubscriptionChannelName({ input, context }) / getPublishChannelName(event) methods. The shared partial-input getChannelName method is deprecated. Publishers pass the event schema's input type; subscribers, filters and channel resolvers get its output type. The raw event travels in oRPC's JSON format, so Date, Map, Set and BigInt survive, and each receiver parses it once. Options include filterFn, authFn, mirrorChannel, overflowMarker, procedure, configured guards and meta. Channels may be static or use channel: { subscribe: ({ input, context }) => key, publish: (parsedEvent) => key }. Carry publishing tenant keys in the event. A factory namespace prefixes channels, mirrors and backlog keys. Each pub/sub definition subscribes to each channel once and parses each payload once for all local subscribers. Each subscriber then runs its filter in arrival order as it reads; its bounded queue holds events before filtering, so onDrop counts unfiltered events. Abort and drain release a subscription at once, even one that is not being read. A lost subscription is retried with exponential backoff and jitter. safePublish logs a failed publish (pubsub.publish_failed) instead of throwing, as on fnLive. createPublisher is the publish-only half.
defineScopedChangeBus binds createPubSub to one scope rule, for buses that tell everyone in a scope (usually an organization) "this changed". Each bus names a channel prefix; events go to ${prefix}:${scope}. Publishing is safe by default, since buses publish after committed writes. inputSchema defaults to z.object({}). bus.channel lets an fnLive share the bus's events, and bus.channelFor({ organizationId }) returns a scope's full channel name, namespace included, e.g. for backlogKey(...).
import { defineScopedChangeBus, invalidationStream } from 'orpc-fn/live'
const defineChangeBus = defineScopedChangeBus(createPubSub, {
subscribe: (context) => activeOrganizationId(context), // throw to refuse
publish: (event: { organizationId: string }) => event.organizationId
})
export const dealChanges = defineChangeBus({
name: 'deal.subscribe',
prefix: 'deal:change',
eventSchema: z.object({ organizationId: z.string(), dealId: z.string() })
})
invalidationStream fans several streams into one stream of { domain } events, so a client holds one connection and invalidates queries per domain. events optionally forwards fields from a source event, for example to raise an alert. Sources are called with {} and the caller's context. Equal events that queue while the consumer is busy collapse into one. When one source ends (a deploy drain) the stream ends, when one throws the stream throws, and every source is closed either way.
export const subscribeChanges = fn({
name: 'realtime.subscribeChanges',
stream: true,
input: z.object({}),
handler: ({ context, signal }) =>
invalidationStream({
context,
signal,
sources: { deal: dealChanges.subscribe, workItem: workItemChanges.subscribe },
events: { deal: (event) => (event.kind === 'won' ? { dealId: event.dealId } : undefined) }
})
})
On the client, useInvalidationStream from orpc-fn/vue or orpc-fn/react consumes it; its domains map is typed from the route's domains.
Graceful shutdown: await shutdown() stops new subscriptions/publication, cancels initialization/retries, ends existing streams and awaits cleanup. It closes a supplied transport only with ownsTransport: true, and a lazy transport: () => … only if it was ever created; it is never created during or after shutdown. A custom transport whose pending subscribe cannot be cancelled may resolve later; that late subscription is immediately released. drainPubSubSubscribers() remains a synchronous, nonterminal operation. Both are per factory.
Recent event IDs deduplicate replay/live overlap; legacy payloads remain readable. Replay is bounded and TTL-limited, so it is not durable resume. A snapshot may already include a queued delta: reducers must use revisions or idempotence. See the revision example. Parsing ingress, replay staging and subscriber queues are bounded; positive queue/TTL/time options are validated.
Transports implement PubSubTransport (publish(messages, backlog?), readBacklog(channel), subscribe(channel, listener, onLost, { onReconnect }?) and optional stats()):
A subscriber connection that is lost, or cannot restore its channels after a reconnect, is closed and replaced; nothing it still emits is delivered. The Redis transports publish and maintain backlogs with one Lua script, sent by SHA, with a fallback to EVAL when Redis returns NOSCRIPT. Await transport.close() when the app owns its lifecycle.
All three return stats(): { channels, listeners, subscriberConnected } on this instance. With activePubSubSubscriberCount() from createFn that covers pub/sub gauges; a transport opens no connection until the first subscribe, so it can be created eagerly:
const transport = bunRedisTransport(redis)
createFn({ /* … */, pubsub: { transport, ownsTransport: true } })
meter.createObservableGauge('pubsub_subscribed_channels').addCallback((g) => g.observe(transport.stats().channels))
meter.createObservableGauge('pubsub_redis_subscriber_connected').addCallback((g) => g.observe(transport.stats().subscriberConnected ? 1 : 0))
``` The Redis Lua batching adapter supports standalone Redis; Redis Cluster requires compatible key slots and is not promised by this adapter.
`streamLiveSnapshots`, `fnLivePatch` and `throwInitialSnapshotError` are exported for custom streams.
## `orpc-fn/mastra`
Optional peers: `@mastra/core` ^1.51, `@orpc/zod`, `@orpc/json-schema`.
```ts
import { createMastraTool } from 'orpc-fn/mastra'
const tool = createMastraTool(getOrder, { requireApproval: true })
// at run time: requestContext.set('orpcContext', context)
Options: id, description, requireApproval, allowMissingInputSchema, onExecuteFinish, contextKey (default 'orpcContext'). Execution goes through oRPC call, so middleware and guards apply.
The tool takes the procedure's raw input and returns its parsed output, and InferToolInput/InferToolOutput/InferUITools see exactly those types. Mastra validates the input (coercing date strings) but passes the raw value on, and the procedure parses that raw value: transforms/refinements must be pure: SDK validation and procedure execution may run them more than once, always on raw input. Agent tools accept objects and unions of object shapes. Other inputs and streams are rejected. Finish-hook failures preserve execution outcomes.
orpc-fn/mcpOptional peers: @orpc/zod, @orpc/json-schema.
Inspection definitions from listTools use Standard Schema and are not McpServer.registerTool configs. Executable registration is a separate optional entry point per SDK major: orpc-fn/mcp/sdk for @modelcontextprotocol/sdk ^1.32.0 and orpc-fn/mcp/server for @modelcontextprotocol/server ^2.0.0:
import { Server } from '@modelcontextprotocol/sdk/server/index.js'
import { hasTag } from 'orpc-fn/mcp'
import { registerMcpTools } from 'orpc-fn/mcp/sdk'
const server = new Server({ name: 'api', version: '1.0.0' }, { capabilities: { tools: {} } })
registerMcpTools(server, router, {
filter: hasTag('external'),
context: ({ signal }) => authenticateToolRequest(signal)
})
// Connect your SDK transport; see examples/adoption/mcp.ts.
For MCP SDK v2 (@modelcontextprotocol/server ^2.0.0), import the same registerMcpTools from orpc-fn/mcp/server instead; options and behaviour are identical:
import { McpServer } from '@modelcontextprotocol/server'
import { registerMcpTools } from 'orpc-fn/mcp/server'
const mcp = new McpServer({ name: 'api', version: '1.0.0' }, { capabilities: { tools: {} } })
registerMcpTools(mcp.server, router, { filter: hasTag('external'), context })
The adapter installs the SDK list/call handlers on a low-level Server (also accessible as McpServer.server). Use one registration per server; these handlers own its tool list. It forwards cancellation, validates/coerces arguments, applies procedure authorization, and formats results as JSON/text MCP content. Override formatResult for images, structured content or other presentation. Tool selection is required. Object unions are supported; streams are rejected. Resolve lazy routers first with oRPC's unlazyRouter.
Sanitized tool-name collisions between distinct procedures throw; identical aliases are deduplicated. Override name to resolve deliberate collisions. Tool names are the sanitized fn names (user.me becomes user-me). Input schemas carry the same JSON Schema as your OpenAPI docs (dates become string/date-time), and validation turns them back into Dates. Validation returns the raw arguments, and call(procedure, args, { context }) parses those: transforms/refinements must be pure: SDK validation and procedure execution may run them more than once, always on raw input. readOnlyHint defaults to GET routes; pass readOnly to change it. hasTag(tag) also works as an oRPC OpenAPIHandler filter.
orpc-fn/cliUses hanzio/cli, which orpc-fn depends on (hanzio ^3.3.0). Your defineCli must come from the same hanzio copy, so keep your own hanzio on a matching major.
import { defineCli } from 'hanzio/cli'
import { procedureCommands } from 'orpc-fn/cli'
const updateDeal = fn({
name: 'deal.update',
summary: 'Move a deal to another stage', // the command's help
meta: { cli: true },
input: z.object({ id: z.string().meta({ positional: true }), stage: z.string() }),
handler: async ({ input, context }) => { /* ... */ }
})
await defineCli({
name: 'app',
globals: z.object({ as: z.string().optional() }),
context: async ({ as }) => ({ user: as ? await loadUser(as) : undefined }),
commands: procedureCommands(router)
}).run() // app deal update d_1 --stage won --as [email protected]
procedureCommands(router) turns the routes declared with meta: { cli: true } into commands nested like the router, kebab-cased (dealRoom.update is deal-room update); add cli?: boolean to your meta type, or select routes with filter. Flags and positionals come from the route's z.object() input. A command calls the procedure through oRPC with the CLI's context, so middleware, guards and tracing run as over HTTP; context: (cliContext, meta) => ... maps that context per procedure and can refuse with a CliError. A 4xx ORPCError prints as CODE: message; anything else keeps its stack. Streams and non-object inputs throw when the commands are built. The procedure parses the already-validated flags again, so transforms must accept their own output.
orpc-fn/honoOptional peers: hono, @orpc/openapi, @orpc/zod, @orpc/json-schema.
import { mountOrpc, normalizeExpoOrigin } from 'orpc-fn/hono'
mountOrpc(app, {
router,
rpcPrefix: '/rpc',
openapi: { prefix: '/api', info: { title: 'API', version: '1.0.0' } },
context: async (c, { headers, timing }) => ({ headers, timing, session: await getSession(headers) }),
normalizeHeaders: normalizeExpoOrigin,
onError: (error, request) => errorTracker.capture(error, request)
})
Behaviour:
Server-Timing header built from the request timing collector: app-recorded queue/auth/context metrics plus library-owned procedure time. Nested calls do not overwrite parent timing; batches aggregate their top-level procedure durations (which may overlap). fn does not mutate an arbitrary context.timing field. formatServerTiming and finishOrpcHeaders are available for composed Fetch adapters.BatchHandlerPlugin).compress: false turns it off). A streaming batch sends each result as it finishes, so it is flushed after every result instead of buffered; Hono's compress() skips event streams entirely. Live event streams are never compressed. A response that already has a content-encoding is left alone.onError sees only unexpected errors.openapi.filter, openapi.smartCoercion and openapi.spec (for security schemes) cover an external API with its own auth.markRequestArrival(request) at the front door (Bun.serve({ fetch }), before routing) makes queue_ms the time until the procedure runs: in-process queueing no handler timer sees. Pass queueMs to measure it another way.context builds the router's context and is checked against it: it is required when { headers, timing } alone does not satisfy the router. It may return a Response, for example a 401.orpc-fn/clientHas no server code, so it is safe in frontends (Vue, React, Expo).
import { createORPCClient } from '@orpc/client'
import { createTanstackQueryUtils } from '@orpc/tanstack-query'
import { createRpcLink, hasOrpcErrorCode } from 'orpc-fn/client'
import type { RouterClient } from '@orpc/server'
import type { router } from '@app/backend' // type-only import
const link = createRpcLink({
url: `${apiOrigin}/rpc`,
fetch: (request, init) => fetch(request, { ...init, credentials: 'include', cache: 'no-store' })
})
export const client: RouterClient<typeof router> = createORPCClient(link)
export const orpc = createTanstackQueryUtils(client) // Vue Query or React Query
if (hasOrpcErrorCode(error, 'NOT_FOUND')) { /* … */ }
toast(userFacingErrorMessage(error, t('errors.generic'))) // server text only when it opts in
Route error messages are developer text, so userFacingErrorMessage shows your localized fallback unless the server opted in with new ORPCError('BAD_REQUEST', { message, data: { userMessage } }). queryCacheSerializer is a serialize/deserialize pair for TanStack Query persisters that round-trips Dates and the other types oRPC sends.
createRpcLink takes every RPCLink option. It also batches parallel calls into one request, matching mountOrpc's server-side batching (batch: { maxSize, exclude, groups }, or false to turn it off). A batch is one request with one client context, so by default only calls without a client context are batched; pass groups to batch calls that share one. A streaming response cannot be batched, and the request doesn't say whether the response will stream, so pass streamPaths generated from the actual router with createStreamManifest (or createStreamManifestAsync for lazy routers). Serialize it to a frontend module; importing that data needs no server runtime. The orpc-fn CLI does that, resolving lazy routers, sorting paths and writing only on change:
orpc-fn stream-manifest src/router.ts#appRouter --out ../web/src/stream-paths.ts # export const streamPaths = [...]
orpc-fn stream-manifest src/router.ts#appRouter --out ../web/src/stream-paths.ts --check # CI: exit 1 when stale
--out x.json writes JSON instead, and --name renames the export. The CLI imports the router, so that module's side effects (database clients, SDKs) still run; it exits when done rather than waiting on their open handles. Run TypeScript routers with bun, tsx or Node ≥ 22.18. Mark custom iterator routes with stream: true or use oRPC's eventIterator output schema. Generated live/pubsub routes are marked automatically. batch.exclude remains an escape hatch, and the legacy subscribe naming heuristic remains as a fallback.
orpc-fn/expoExpo itself is passed in, not imported. The fetch must resolve to a response with the members oRPC reads (expo/fetch does).
import { fetch } from 'expo/fetch'
import { Platform } from 'react-native'
import { createBetterAuthExpoLink } from 'orpc-fn/expo'
const link = createBetterAuthExpoLink({
url: `${baseUrl}/api/rpc`,
fetch,
native: Platform.OS !== 'web',
getCookie: () => authClient.getCookie(), // Better Auth Expo
getExpoOrigin,
headers: () => ({ 'x-orpc-source': 'expo-react' })
})
On native, calls go through expo/fetch, the only fetch that streams there. In Expo SDK 55, cancelling a stream makes expo/fetch close its body twice and throw Invalid state: Controller is already closed; patch src/winter/fetch/FetchResponse.ts so its didComplete and didFailWithError listeners return when isControllerClosed is already set (with bun patch expo or patch-package). The bridge forwards the request body and the abort signal. Without the signal, cancelled subscriptions keep their sockets open, the device's HTTP pool runs out, and later requests hang. Cookies are sent in a header (credentials: 'omit'), together with expo-origin and x-skip-oauth-proxy, the same headers @better-auth/expo sends. On the server, mountOrpc({ normalizeHeaders: normalizeExpoOrigin }) copies expo-origin to origin. The generic createExpoLink forwards upstream headers (including async/context-aware resolvers) and adds no provider-specific headers. createBetterAuthExpoLink supplies the cookie/origin preset above. On web, the link uses the platform fetch with cookies, or your explicit webFetch override. createExpoFetch is exported for a custom link.
orpc-fn/vuePeers: vue, @tanstack/vue-query.
// main.ts
app.use(VueQueryPlugin, { queryClient })
provideOrpc(app, { client, utils: orpc })
declare module 'orpc-fn/vue' {
interface Register { client: typeof client; utils: typeof orpc }
}
const { useQuery, useMutation, invalidate } = useApiClient()
const orders = useQuery((o) => o.order.list.queryOptions({ input: { search: search.value } }))
const save = useMutation((o) => o.order.update.mutationOptions({ onSuccess: () => invalidate((o) => o.order.key()) }))
// Form state: `inputData` is typed from the route; `initialInput` is checked against it.
const { inputData: form, mutate } = useMutation((o) => o.order.update.mutationOptions(), {
initialInput: { id: props.id, data: {} }
})
// A helper that takes the composables: `ApiClient` is bound to the registered utils.
function invalidateOrders(invalidate: ApiClient['invalidate']) {
void invalidate((o) => o.order.key())
}
// The client half of `fnLive`: HTTP seeds the cache, the stream keeps it current.
const order = useLiveQuery(
() => ({ id: props.id }),
(input, o) => o.order.get.queryOptions({ input }),
(o) => o.order.watch.call // the composable passes the input and an abort signal
)
// Once, in the signed-in shell: one stream of `{ domain }` events invalidates list queries.
useInvalidationStream({}, (o) => o.realtime.subscribeChanges.call, {
domains: { order: { debounceMs: 100, queryKeys: [orpc.order.key()] } },
logger
})
useApiClient() also returns the utils as orpc, for calls outside a composable such as queryClient.fetchQuery(orpc.unit.list.queryOptions({ input })). useOrpcSubscription (also useApiClient().useSubscription) reconnects with jittered backoff and closes while the page is hidden (respectVisibility: false keeps it). useLiveQuery writes each stream event under the key it was opened for, so an event in flight from a previous input is dropped. A stream whose events are not the cached type (an fnLive patch route, or a selected cache) does not compile without applyEvent, and the stream's input must match the query's unless you pass subscriptionInput. Use mergeSeed when the seed can resolve after stream events. For large live payloads such as chat timelines, pass vue-query's shallow: true in the query options so the data is not made deeply reactive; mutate it only through setQueryData. Live queries carry meta.live, so invalidation streams skip them. Leave refetchType at TanStack's 'active': 'all' refetches every cached query of a domain on each event, including screens nobody has open, which refetch on their next visit anyway. useSubscriptionInvalidates treats a stream as a "something changed" ping, and useQueryErrorMessage is userFacingErrorMessage as a computed. A surface that talks to another API passes orpc to each composable instead of using provideOrpc.
orpc-fn/vue/testing runs composables in a real app without a DOM: withOrpcApp(setup, { utils }), createTestStream() for a subscription the test drives (await stream.push(event) resolves once it was handled), and createOrpcMock({ fn: mock }) for a deep stand-in of the utils.
orpc-fn/reactPeers: react, @tanstack/react-query. Works on React Native: pausing follows TanStack's focusManager, so wire it to AppState once, as TanStack recommends:
focusManager.setEventListener((setFocused) => {
const subscription = AppState.addEventListener('change', (state) => setFocused(state === 'active'))
return () => subscription.remove()
})
const { status } = useOrpcSubscription({ id }, client.order.watch, { onEvent: (order) => setOrder(order) })
// Same contract as the Vue `useLiveQuery`, with the client procedure passed as is.
// For a large cache your `applyEvent` keeps immutable, pass `structuralSharing: false` in the query options.
const order = useLiveQuery({ id }, (input) => orpc.order.get.queryOptions({ input }), client.order.watch)
useInvalidationStream({}, client.realtime.subscribeChanges, {
enabled: signedIn,
domains: { order: { debounceMs: 500, queryKeys: [orpc.order.list.key()] } }
})
// Persist the cache per user: the buster drops someone else's saved cache, the hook clears memory.
const owner = session ? `${session.user.id}:${session.activeOrganizationId ?? ''}` : null
useQueryCacheOwner(owner)
const persister = createAsyncStoragePersister({ storage: AsyncStorage, ...queryCacheSerializer })
<PersistQueryClientProvider client={queryClient} persistOptions={{ persister, buster: owner ?? '' }} />
MIT
Define a backend function once: a typed oRPC procedure with tracing, logging, guards, injected handler extras, nested calls and live queries over Redis.
{ "mcpServers": { "orpc-fn": { "command": "npx", "args": [ "-y", "[email protected]" ] } }}टूल
अभी टूल सूची नहीं है। रिमोट सर्वर को पहले हेल्थ चेक के बाद सूची मिलती है।
FAQ
इस पेज पर Cursor, Claude Desktop, Claude Code या VS Code के लिए stdio कॉन्फ़िग कॉपी करें। यह पब्लिशर का इंस्टॉल इस्तेमाल करता है, PlugMCP प्रॉक्सी नहीं।
इस रिकॉर्ड में कोई ज़रूरी एनवायरनमेंट वेरिएबल दर्ज नहीं है।
Orpc Fn stdio पर लोकल चलता है।
पिछली जाँच 2026-11-13 को हुई। Define a backend function once: a typed oRPC procedure with tracing, logging, guards, injected handler extras, nested calls and live queries over Redis.