Can one MCP endpoint cover Gmail, Notion, Jira and other sources?

Yes. Instead of enabling a separate connector per tool and carrying all of them in every prompt, you point the model at one endpoint that serves the sources behind it. You ask once — 'what's the status of the launch?' — and it pulls from the source that holds the answer, rather than you naming the tool up front.

How does the endpoint know which source to use?

It routes on the intent of the request. A question about a message goes to mail, a question about a task goes to your tracker, a question about a doc goes to your workspace. The model isn't reasoning over ten tool definitions to pick one; the routing happens behind the single endpoint, so the answer comes back leaner and with less room to mis-route.

Does a single endpoint mean one login for everything?

It means one endpoint, backed by your own grants for each source. You still authorise each connection with the source's own permissions — the endpoint serves those connectors, it doesn't replace or bypass their auth. Stash Lite is built on Stash, with one Google sign-in across both.

One MCP endpoint for Gmail, Notion, Jira and more

Q: Is my email or data stored to make this work?

No. For connector reads, responses are reshaped as they pass through and nothing is retained. For email specifically, we never store your messages — the mailbox connection runs on your own OAuth grant. The data passes through to the AI; it does not live in Stash Lite.

What is a single MCP endpoint?

Normally each tool is its own MCP connector. Connect Gmail, Notion and Jira and the model carries three sets of tool definitions in the context of every call — plus whatever else you've wired in. That baseline rides along on every turn, and a long list of tools gives each request more places to get routed wrong. We covered that cost in the Claude MCP connector cost spiral.

A single endpoint collapses that. The model sees one route in. The work of talking to each source, and picking the fields that matter, happens behind the endpoint instead of in the model's context. One call, not ten.

"Ask once" — how intent routing picks the source

The endpoint routes on what you're asking for, not on you naming a tool. A question about a message is a mail question. A question about a ticket's status is a tracker question. A question about a spec is a workspace question. The routing reads that intent and pulls from the source that holds the answer.

The practical difference is that you stop being the router. Instead of "use the Jira tool to check the launch issue," you ask "what's blocking the launch?" and the right source answers. Fewer tools in context also means fewer chances for the model to reach for the wrong one — a quieter, leaner call.

Does one endpoint mean one login for everything?

No, and it shouldn't. One endpoint is about how the model reaches your sources, not about collapsing their permissions. Each source is still connected with its own grant and its own scope — the endpoint serves those connectors, it doesn't replace or bypass their auth. You keep the same access controls you already have.

Stash Lite is built on Stash, so there's one Google sign-in across both products. That's a single sign-in for Stash Lite itself, not a master key to your tools. Your Gmail, Notion and Jira access stays governed by their own grants.

What about email — is it stored?

This is worth being precise about, because the honest answer depends on the source. For connector reads, responses are reshaped as they pass through and nothing is retained — the document passes through to the AI and does not live in Stash Lite. For email specifically, we never store your messages; the mailbox connection runs on your own OAuth grant so the model can read what it needs in the moment, without a copy being kept.

The reason to keep those two statements separate rather than blur them into one slogan: they're true for different reasons, and a security review will ask. Same principle, though — read-only where it can be, and no standing copy of your content.

Keeping the answer correct when one call spans sources

A fair worry with routing is whether "ask once" quietly drops something. It shouldn't. The endpoint reshapes responses to send the model the fields an answer needs — it isn't giving the model less to reason with, it's not shipping the parts that were never going to be used. If a question genuinely needs the heavy fields, those are kept for that response. We wrote up how that line is drawn in how reshaping rules are written.

So the multi-source call stays as accurate as asking each connector directly — same data behind it, same auth — just leaner and with one route in.

Try it

Stash Lite serves the connectors you already use — Gmail alongside Notion, Coda, Jira and Confluence, HubSpot, GitHub, Salesforce, and more — behind one endpoint, and reshapes each response in transit so the call stays lean. From our preliminary, directional testing, heavy reads can drop up to about 80% of their token cost through a reshaping layer; that's a ceiling on favourable workloads, not a promise, and worth measuring on your own traffic. It's in early access, so it's a new service and pricing may change — you can leave any time.

See the single-endpoint setup and join the early-access list: stashlite.com/consolidate.

This is part of our Stash Lite series on the real mechanics of MCP cost at scale, measured carefully and hedged where it should be. Related: how to consolidate multiple MCP connectors into one call and the Claude MCP connector cost spiral.