One MCP to rule them all. A single MCP endpoint that aggregates many downstream connectors — remote MCP servers and plain HTTP APIs — and presents agents a fixed set of nine meta-tools instead of hundreds of individual tools.
┌── remoteMcp("notion") → mcp.notion.com
Claude / Cursor ── MCP ──▶ connecta ───┼── remoteMcp("linear") → mcp.linear.app
sees 9 tools /mcp ├── api("resend") → fetch(...)
└── api("internal") → fetch(...)
The agent always sees exactly nine tools; it searches when it needs something and describes only what it will call. Progressive disclosure is the point. Connectors are declared in TypeScript (config as code — adding one is a code change + deploy, no database of integrations), and the same fetch-based core runs on Node and Cloudflare Workers. Optionally, code mode adds a tenth tool that runs model-written orchestration code in a sandbox.
Node deployments require Node.js 20.9 or newer.
npm install @zackbart/connectaProvider and runtime integrations remain consumer-owned. Install only the optional packages a deployment uses—for example:
npm install @clerk/backend # optional Clerk auth adapter
npm install quickjs-emscripten # optional Node code-mode executor
npm install @cloudflare/codemode # optional Worker code-mode executorA tool address is <connectorId>.<toolName> (e.g. notion.search).
| Tool | Input | Returns |
|---|---|---|
list_connectors |
{ probe? } |
live (probe: true, default) or cached health, tool count, and recent real-call observations |
skills |
{ name? } |
lists or fetches the concise usage guide for choosing among the meta-tools |
search_tools |
{ query?, connector?, limit?, offset?, fullDescriptions?, includeSchemas? } |
ranked, paginated matches; optionally includes compact/raw schemas to remove a round trip |
describe_tools |
{ addresses[], format?, fullDescriptions? } |
names, descriptions, input/output schemas, and behavior annotations |
call_tool |
{ address, args?, fields?, resultMode?, timeoutMs?, maxRetries?, diagnostics? } |
invokes only tools explicitly annotated readOnlyHint: true |
call_destructive_tool |
same as call_tool |
invokes unannotated, write-capable, or destructive tools through a host-visible approval boundary |
authorize_connector |
{ connector, force? } |
starts (or with force, restarts) the downstream OAuth flow; returns the authorizationUrl to open |
get_result |
{ id, offset?, maxBytes? } |
a byte-slice page of a truncated result — { text, offset, nextOffset?, totalBytes } |
batch_call |
{ calls, resultMode?, timeoutMs?, maxRetries? } |
1–10 parallel calls sharing request-scoped clients, with attempts/timing/errors |
execute_code (optional) |
{ code } |
result + logs of bounded async JS orchestration over explicitly read-only tools; registered only when an executor is configured |
This repository owns the reusable package. Each deployment should be a small,
separate Worker project that pins an exact package version and owns only its
connector configuration, auth policy, domain, bindings, migrations, and secrets.
examples/worker/ is the starting template.
import { DynamicWorkerExecutor } from "@cloudflare/codemode";
import { bearerToken, createConnecta, remoteMcp } from "@zackbart/connecta";
import { clerkAuth } from "@zackbart/connecta/auth/clerk";
import { cloudflareKvStorage } from "./cloudflare-kv.js";
const build = (env: Env) =>
createConnecta({
publicUrl: env.PUBLIC_URL,
storage: cloudflareKvStorage(env.CONNECTA_KV),
// Code mode (optional): needs a `worker_loaders` binding in wrangler.jsonc.
executor: new DynamicWorkerExecutor({ loader: env.LOADER }),
auth: [
bearerToken(env.CONNECTA_TOKEN),
clerkAuth({
publishableKey: env.CLERK_PUBLISHABLE_KEY,
secretKey: env.CLERK_SECRET_KEY,
publicUrl: env.PUBLIC_URL,
}),
],
connectors: [
remoteMcp("notion", {
url: "https://mcp.notion.com/mcp",
auth: { type: "headers", headers: { Authorization: `Bearer ${env.NOTION_TOKEN}` } },
}),
],
});
// Lazy per-isolate singleton: keeps only serializable tool/catalog data warm.
let connecta: ReturnType<typeof build> | undefined;
export default {
fetch(request: Request, env: Env): Promise<Response> {
connecta ??= build(env);
return connecta.fetch(request);
},
};Deployable example: examples/worker/.
Its Cloudflare KV and D1 implementations are deployment-owned examples over the
generic KVStorage and ActivityStore contracts; they are not package exports.
clerkAuth is an optional adapter. Install @clerk/backend and import it from
@zackbart/connecta/auth/clerk only in deployments that use Clerk. Other
identity providers can implement the exported InboundAuth interface.
import { api, bearerToken, createConnecta } from "@zackbart/connecta";
import { fileStorage, listen } from "@zackbart/connecta/node";
import { quickJsExecutor } from "@zackbart/connecta/quickjs";
const connecta = createConnecta({
storage: fileStorage("./.connecta-state.json"), // or memoryStorage()
auth: bearerToken(process.env.CONNECTA_TOKEN!),
executor: quickJsExecutor(), // code mode (optional): QuickJS/WASM sandbox
connectors: [
api("time", {
description: "Time — clock utilities",
tools: [{
name: "get_now",
description: "Return the current ISO timestamp.",
inputSchema: { type: "object", properties: {} },
annotations: { readOnlyHint: true },
handler: async () => ({ now: new Date().toISOString() }),
}],
}),
],
});
listen(connecta, 8787); // http://localhost:8787/mcpExample: examples/node/. Docker (single-service compose
stack): examples/docker/.
With an executor configured, the model can write an async function instead of
making one call_tool round trip per step — loops, joins across connectors,
filtering big responses down in-sandbox before they hit the context window:
async () => {
const pages = await notion.search({ query: "roadmap" });
return pages.results.map((p) => p.title);
}The sandbox has no network/filesystem/env — only explicitly read-only
connector globals and
connecta.call, connecta.batch, connecta.search, and connecta.describe;
credentials stay host-side. A run may make at most 20 host calls, a
connecta.batch may contain at most 10 calls, and every host call has a
15-second deadline. Executors:
DynamicWorkerExecutor (@cloudflare/codemode,
Workers, Worker Loader binding — open beta) or quickJsExecutor()
(@zackbart/connecta/quickjs, QuickJS-in-WASM, runs anywhere). Omit the executor and
connecta is exactly the nine-tool server. Details:
docs §13.
API connectors can declare one operator-managed credential or a named set of
credential fields. Connecta renders
Add / Replace / Test / Remove controls inside that connector's /ui card,
encrypts the values with AES-GCM, and stores only ciphertext in the deployment's
existing KVStorage. Credentials are available only to the connector through
ctx.credential.get(), get(name), or getAll(); they are never returned by
/ui, MCP tools, or code mode.
api("example", {
description: "Example — authenticated API",
credential: {
label: "API token",
description: "Token used for outbound Example API requests.",
},
tools: [{
name: "get_profile",
description: "Get the authenticated Example profile.",
inputSchema: { type: "object", properties: {} },
annotations: { readOnlyHint: true },
handler: async (_args, ctx) => {
const token = await ctx.credential?.get();
if (!token) throw new Error("Example API token is not configured.");
return fetch("https://api.example.com/profile", {
headers: { Authorization: `Bearer ${token}` },
}).then((response) => response.json());
},
}],
})Set credentialEncryptionKey on createConnecta to a base64-encoded 32-byte
key held in the runtime's secret store (openssl rand -base64 32) — a connector
that declares a credential without one fails at construction rather than booting
with an unusable vault. Credential mutation routes require the configured Clerk
provider and a same-origin browser request; the static inbound bearer cannot
administer the vault.
Connecta does not bundle service-specific HTTP API connectors. Package consumers
define them with api() (or implement Connector directly), keeping endpoint,
credential, and tool choices in the consuming project.
Connecta can record which resolved downstream tools were actually invoked
without storing their arguments, results, generated code, search text, or raw
errors. Supply a vendor-neutral activity store:
const events: ToolCallActivityEvent[] = [];
const connecta = createConnecta({
connectors,
activity: {
record(event) {
events.push(event);
},
async list({ limit }) {
return { events: events.slice(-limit).reverse() };
},
},
});One final event is emitted for each resolved connector call made through
call_tool, call_destructive_tool, batch_call, or execute_code; retries
remain one event with an attempts count. Implementing list enables the
authenticated Activity tab in /ui; activityReadGate can narrow reads
further. Writes are best-effort and never change a tool result. Clerk calls
carry the Clerk user ID; shared bearer calls are honestly labeled as bearer
unless bearerToken(secret, { subjectId }) assigns that credential a stable
subject. On Workers, pass ctx through to connecta.fetch(request, env, ctx)
so async writes settle on waitUntil.
examples/worker/src/d1-activity.ts is
a complete D1 implementation with keyset paging and a retention pass.
GET /ui is a read-only dashboard with no build step: connector health, tool
counts and descriptions with a client-side filter, downstream authorization
links, the credential controls above, and an Activity tab when an activity store
is configured. The shell is open because it carries no data; everything it shows
comes from /ui/data, behind the same auth gate as /mcp. With Clerk
configured it signs operators in through Clerk's hosted portal; a bearer-only
deployment falls back to a pasted token.
Every deployment-facing label and mark on /ui and the OAuth result pages comes
from ConnectaConfig.branding — productName, ownerName, their URLs,
description, pageTitle, themeColor, and favicon — each falling back to a
neutral Connecta default. Nothing about the operator is baked into the package.
docs/documentation.md— how everything works: architecture, the meta-tools, connectors, inbound auth, downstream OAuth, storage, running it (Node / Workers / Docker), Clerk setup, testing, troubleshooting, code mode, the status UI, and activity history.docs/design.md— why it's built this way, and the non-goals.CHANGELOG.md— what changed in each release.
