Gateways and worker fleets
A gateway — a proxy, a worker pool, any process that fronts one MCP server with many short-lived clients — probes the server once and reuses the answer for every connection after it.
Connect with a prior discover result
connect() takes an optional prior: a cached era verdict (PriorDiscovery). Its modern arm, { kind: 'modern', discover }, wraps a persisted DiscoverResult from an earlier probe — connect() adopts the server's advertisement directly and sends nothing on the wire.
import { Client, StreamableHTTPClientTransport } from '@modelcontextprotocol/client';
const url = new URL('http://localhost:3000/mcp');
// Probe once …
const bootstrap = new Client({ name: 'gateway', version: '1.0.0' }, { versionNegotiation: { mode: 'auto' } });
await bootstrap.connect(new StreamableHTTPClientTransport(url));
const persisted = JSON.stringify(bootstrap.getDiscoverResult());
// … then every other client connects with zero round trips.
const worker = new Client({ name: 'worker', version: '1.0.0' });
await worker.connect(new StreamableHTTPClientTransport(url), { prior: { kind: 'modern', discover: JSON.parse(persisted) } });
worker is connected: callTool works immediately, and the server has not heard from it yet.
The modern verdict is 2026-07-28+ only — see Protocol versions. For a server known to be legacy, supply the negative verdict instead — see Skip the probe for a known-legacy server.
Probe once at bootstrap
An 'auto'-mode (or pinned) connect sends server/discover and records the answer; getDiscoverResult() reads it back.
console.log(bootstrap.getDiscoverResult());
The recorded value is the server's whole advertisement — supported versions, capabilities, identity, instructions:
{
ttlMs: 0,
cacheScope: 'private',
supportedVersions: [ '2026-07-28' ],
capabilities: { tools: { listChanged: true } },
serverInfo: { name: 'gateway-target', version: '1.0.0' },
resultType: 'complete'
}
::: tip
A client on a modern-era connection can re-probe at any time: await client.discover() sends server/discover and updates getDiscoverResult(). On a legacy-era connection discover() throws (server/discover is not a 2025-era method) — to re-check a legacy verdict, reconnect without a prior under mode: 'auto', as in Caching discovery verdicts. A default-mode connect never probes, so its getDiscoverResult() is undefined — Protocol versions lists the negotiation modes.
:::
Persist the advertisement
The value is plain JSON. Write the string to Redis, a config map, or a process-local cache; parse it back and wrap it as the modern verdict wherever a client needs it.
import type { DiscoverResult, PriorDiscovery } from '@modelcontextprotocol/client';
const discover = JSON.parse(persisted) as DiscoverResult;
const prior: PriorDiscovery = { kind: 'modern', discover };
Nothing about prior is tied to the process that probed: any client that can reach the same URL can adopt it.
Fan out to workers
Build every replica from the same blob; the request_count call after them is the proof.
const fleet = await Promise.all(
['worker-a', 'worker-b', 'worker-c'].map(async name => {
const replica = new Client({ name, version: '1.0.0' });
await replica.connect(new StreamableHTTPClientTransport(url), { prior });
return replica;
})
);
const proof = await worker.callTool({ name: 'request_count' });
console.log(proof.structuredContent);
request_count is a tool on this page's example server that returns how many MCP requests reached the process. Five clients are connected by now — bootstrap, worker, three replicas — and the server has answered two requests:
{ requests: 2 }
The bootstrap probe was the first request and the request_count call itself the second. The four connect({ prior }) calls sent nothing.
Reuse only within one authorization context
The advertisement is what the server returned to the credential that probed.
::: warning
Never share a persisted DiscoverResult across principals — key the blob on the authorization context that obtained it (a credential hash works). The server still authorizes every request, so a wider prior grants nothing, but it misleads client-side capability gating.
:::
Open a listen stream when a worker needs notifications
A modern-verdict connect({ prior }) never auto-opens a subscriptions/listen stream — the client is request-only until you open one yourself. (A legacy-verdict connect is an ordinary 2025-era connection: unsolicited notifications, no listen().)
const subscription = await worker.listen({ toolsListChanged: true });
console.log(subscription.honoredFilter);
The server acknowledges the filter it agreed to honor:
{ toolsListChanged: true }
From here the stream behaves like any other subscription — Subscriptions covers the notification handlers and the close semantics.
::: info
A listChanged option configured on a modern-verdict client registers its handlers but stays silent: no stream opens until you call listen().
:::
Handle a stale or incompatible advertisement
A modern verdict whose discover shares no 2026-07-28+ revision with the client rejects with SdkError(EraNegotiationFailed) before anything reaches the transport.
import { SdkError, SdkErrorCode } from '@modelcontextprotocol/client';
const stale: DiscoverResult = { ...discover, supportedVersions: ['2025-06-18'] };
const late = new Client({ name: 'worker-d', version: '1.0.0' }, { versionNegotiation: { mode: 'auto' } });
try {
await late.connect(new StreamableHTTPClientTransport(url), { prior: { kind: 'modern', discover: stale } });
} catch (error) {
if (!(error instanceof SdkError) || error.code !== SdkErrorCode.EraNegotiationFailed) throw error;
console.log(error.code);
// Fall back to a fresh probe, then re-persist getDiscoverResult().
await late.connect(new StreamableHTTPClientTransport(url));
console.log('re-probed:', late.getNegotiatedProtocolVersion());
}
The rejection happens before the transport starts, so the same Client connects again on the fallback path:
ERA_NEGOTIATION_FAILED
re-probed: 2026-07-28
Replace the persisted blob with the fresh getDiscoverResult() and the rest of the fleet recovers on its next read.
Skip the probe for a known-legacy server
When out-of-band metadata already says the server is pre-2026 — a registry entry, an earlier connection's outcome — an 'auto'-mode probe is a round trip that fails on every single connect. Supply the negative verdict instead: PriorDiscovery's { kind: 'legacy' } arm skips the probe and goes straight to the initialize handshake.
const pinnedLegacy = new Client({ name: 'worker-e', version: '1.0.0' }, { versionNegotiation: { mode: 'auto' } });
await pinnedLegacy.connect(new StreamableHTTPClientTransport(url), {
prior: { kind: 'legacy' }
});
console.log(pinnedLegacy.getProtocolEra());
legacy
Freshness is your responsibility: the SDK adopts whatever verdict you hand it. A stale modern verdict fails loudly at the first request, but a stale legacy verdict succeeds silently forever — an upgraded server still answers initialize, so nothing ever corrects it. Date cached legacy verdicts in your own storage and stop supplying them past your policy horizon; with no prior, the configured mode decides again (under mode: 'auto' the connect re-probes, so the upgrade is discovered).
Caching discovery verdicts
The pieces above compose into the full host-side loop: probe once, cache the verdict under your own timestamp, and gate every later connect on a freshness check you control.
The first connect under mode: 'auto' pays one probe. Afterwards the outcome is readable on the client: getDiscoverResult() returns the DiscoverResult on a modern server, and on a connected client its absence means the era is legacy. Store that verdict together with when you stored it — the Map below keeps storedAt explicitly; in a real deployment the store does the dating for you (a Redis key TTL, where an expired read simply comes back empty, or a database row's created_at column).
Before each connect, supply the cached verdict only while your own policy says it is fresh — a fixed horizon here, any predicate in practice. Supplying undefined under mode: 'auto' is the re-probe, and the fresh outcome re-populates the cache. The timestamp matters most for the legacy branch: a stale legacy verdict succeeds silently forever (an upgraded server still answers initialize), so only the timestamp ever retires it; a stale modern verdict fails loudly at the first request.
// Host-side verdict cache. A real deployment keeps this in Redis (a key TTL
// does the dating: an expired read is no prior) or a database row with a
// created_at column; the Map stores the timestamp explicitly.
const verdicts = new Map<string, { verdict: PriorDiscovery; storedAt: number }>();
const HORIZON_MS = 24 * 60 * 60 * 1000; // fixed-horizon freshness policy
const fresh = (entry: { storedAt: number }): boolean => Date.now() - entry.storedAt < HORIZON_MS;
// Count server/discover probes at the fetch layer (the wire trace).
let probes = 0;
const tracingFetch: typeof fetch = async (input, init) => {
if (typeof init?.body === 'string' && init.body.includes('"server/discover"')) probes++;
return fetch(input, init);
};
async function connectCached(key: string): Promise<Client> {
const entry = verdicts.get(key);
// An entry past the horizon is not supplied — under mode: 'auto' that IS
// the re-probe, and the fresh outcome below re-populates the cache.
const prior = entry && fresh(entry) ? entry.verdict : undefined;
const client = new Client({ name: 'cached-worker', version: '1.0.0' }, { versionNegotiation: { mode: 'auto' } });
await client.connect(new StreamableHTTPClientTransport(url, { fetch: tracingFetch }), { prior });
if (prior === undefined) {
// Fresh outcome: getDiscoverResult() is the DiscoverResult on a modern
// server; its absence on a connected client means the era is legacy.
const discover = client.getDiscoverResult();
// Date the entry with the host's own clock — only this timestamp ever
// retires a legacy verdict (a stale one succeeds silently forever).
verdicts.set(key, {
verdict: discover ? { kind: 'modern', discover } : { kind: 'legacy' },
storedAt: Date.now()
});
}
return client;
}
const first = await connectCached('gateway-target'); // no entry: probes
console.log('probes after first connect:', probes);
const second = await connectCached('gateway-target'); // fresh entry: verdict supplied, no probe
console.log('probes after second connect:', probes);
verdicts.get('gateway-target')!.storedAt -= HORIZON_MS + 1; // the horizon passes
const third = await connectCached('gateway-target'); // stale: dropped, re-probed, re-cached
console.log('probes after the horizon passes:', probes);
probes after first connect: 1
probes after second connect: 1
probes after the horizon passes: 2
The wire trace shows the shape of the loop: one probe to fill the cache, none while the verdict is fresh, one more when the horizon forces rediscovery.
Recap
connect(transport, { prior: { kind: 'modern', discover } })adopts a persistedDiscoverResultwith zero round trips.- The advertisement comes from one
'auto'-mode or pinned probe — or an explicitclient.discover()— andgetDiscoverResult()reads it back. - The value is plain JSON: stringify it into a shared cache, parse it back and wrap it as the modern verdict in any process that fronts the same server.
- Reuse a
DiscoverResultonly across clients that present the same authorization context. - Modern-verdict clients are request-only; call
listen()on the one that needs notifications. - An incompatible
priorrejects withSdkError(EraNegotiationFailed); fall back to a fresh probe and re-persist. - A known-legacy server takes
prior: { kind: 'legacy' }— no probe, straight toinitialize. Stale legacy verdicts fail silently (an upgraded server still answersinitialize), so date them in your own storage and stop supplying them past your policy horizon.
Source: docs/advanced/gateway.md