One URL for every MCP server.
An MCP aggregator combines many MCP servers behind a single endpoint, so every agent authenticates once instead of once per service. Gentkey is an aggregator with a policy gate: scoped write grants, hard constraints, and an audit trail for every call.
https://app.gentkey.com/mcpclaude.ai (web & mobile) · Claude Code · Cursor · any MCP clientPer-client, per-service setup doesn’t scale.
Without an aggregator, every agent surface needs every service configured separately: an API key in Claude Code’s config, the same key again in Cursor’s, an OAuth dance per connector in claude.ai — multiplied by every service you use. Keys end up in plaintext config files on every machine, and no one can say which agent did what with them.
On claude.ai’s free plan the arithmetic is starker: you get one custom connector. Pointing it at an aggregator turns one slot into all of them.
Aggregation is table stakes. Governance isn’t.
Writes are opt-in
Combining servers multiplies what an agent can touch. Gentkey gates every state-changing tool behind an explicit, revocable grant — reads work out of the box where curated.
Bounds, enforced with arithmetic
Grants carry hard limits — max $ change per call, max total $ moved per day, max writes per hour. A $500 budget change under a $50-per-call cap is denied before it reaches the vendor.
Credentials stay out of context
Tokens are encrypted at rest and injected server-side at call time. Agents authenticate to Gentkey with OAuth; they never hold a service credential.
Every decision, attributed
Allowed, denied, or denied by constraint — each call is logged with the agent, tool, and reason.
No context-window tax
Past 20 tools, agents get three small meta-tools and search the catalog instead of loading every schema up front.
Any server, same rules
19curated connectors, plus any remote MCP server by URL — upstream tools default to write-gated; vendor self-labeling isn’t trusted.
Fair questions.
What's the difference between an MCP aggregator, a gateway, and a proxy?
A proxy forwards one server's traffic (often bridging local stdio to the network). An aggregator combines many servers behind one endpoint. A gateway is an aggregator that also enforces policy — authentication, permissions, audit. Gentkey is a gateway: aggregation plus scoped grants, constraints, and logging.
Why does the claude.ai free plan make an aggregator almost mandatory?
Free claude.ai accounts currently get a single custom connector. Point that one slot at an aggregator and it fronts every service you've connected — one OAuth flow instead of choosing which single integration you can live with.
Doesn't loading tools from many servers blow up the context window?
It does if the aggregator forwards every schema. Past 20 tools, Gentkey hands agents three small meta-tools and lets them search the catalog on demand instead of loading every schema up front.
When is a hosted aggregator the wrong choice?
If your requirements are all self-hosted — everything on your own infrastructure, local stdio servers, no third-party custody — an open-source self-hosted gateway (MetaMCP, IBM's Context Forge, Docker's MCP Gateway) is the better fit. Gentkey's case is hosted reach (claude.ai web and mobile included), credential custody, and enforced write policy without running your own infrastructure.
What happens to a write the aggregator hasn't been told to allow?
It's denied before it reaches the upstream service, and the denial is logged with the grant that would have authorized it. Reads work out of the box where curated; writes are opt-in, per agent, revocable.
One URL. Smaller keys.
Sign in, connect an account, and hand every agent the same governed endpoint.