100+ integrations are ready to connect to your AI.Connect your first MCP

We hold the credentials. Your AI gets the tools

PopMCP hosts MCP servers for 100+ business tools, so an agent can use what you already pay for.No local process, no server to maintain, and no credential leaving our side of the wire.

Browse the catalog

Every agent demo ends at the same wall

The demo always works. Then you point it at your own store, your own accounting file, your client's ad account, and you are writing an MCP server. One per vendor, each with its own auth, its own token handling, its own rate limits, its own uptime. Ten tools is ten builds and ten things to maintain before anyone has automated anything.

That work is the same for every business that needs it, which makes it a strange thing for every business to do separately. PopMCP is the version where it is already done, and three gaps in particular kept turning up in the same conversation.

  1. One account per integration

    Native connectors in ChatGPT and Claude typically bind to a single account or workspace. Agencies and multi-brand teams run several stores, ad accounts, and client environments at once, and there was nowhere to put the second one.

  2. A thin slice of each tool

    Most official MCP servers expose a small curated set of actions. That is enough to demo and rarely enough to run an operation on, so the work drops back to tabs and copy-paste at exactly the point it gets useful.

  3. Context spent before the work starts

    Loading an entire tool catalog into every conversation burns the context window on tools the task will never call. Keeping frequent tools ready and searching for the rest on demand leaves room for the actual job.

Five decisions you can check for yourself

Every one of these is visible on the pricing page or in the legal center. None of them is a promise about the future.

We do not charge for tool calls

A single conversation can run eight to fifteen calls. Pricing that asks you to predict a monthly number is pricing that charges you more for using the product properly. We charge for connections and workspaces, which you can count before you buy.

Your credentials never reach the AI client

Provider secrets stay with us, encrypted at rest with AES-256-GCM. Claude, ChatGPT and your editor sign in to PopMCP with OAuth and only reach the connectors you approve. Removing access never touches the credential underneath.

More of the catalog, not a curated slice

Most official MCP servers expose a small, safe subset of what their API can do. Where a provider allows it, we expose the broader catalog, because the action you need is usually the one that got cut.

Multiple accounts is the normal case

Agencies run a store and an ad account per client. Most connectors assume one of everything. Separate workspaces, separate connections and separate usage data are how this was built, and per-member access control on every paid plan means a contractor sees one client rather than all of them.

Nothing is deleted the day you downgrade

Drop a plan and anything over the new limit is locked rather than removed, kept for 90 days, and back the moment you subscribe again. On a paid downgrade you choose what to keep. When a trial or subscription ends, your oldest workspace and connections stay live.

Three kinds of team, one endpoint

The shape of the problem changes with the team. What does not change is that it ends in the same place: one hosted URL per connected account.

Agencies

A workspace per client, with separate connections and separate usage data, so one client's numbers never appear in another's report. Per-member workspace access control on every paid plan means a contractor gets the one account they work on.

Ops and revenue teams

The people who live in a storefront, an email platform, a CRM and a spreadsheet, and who would rather an agent did the pulling and the drafting than the tab-switching.

Small dev teams

You could build the connector. The real question is whether you want to own auth, rate limiting, token revocation and uptime for a dozen vendors' APIs for as long as the product exists.

The specifics, not the reassurance

Every line here is in the security overview, the privacy policy or the subprocessor list.

  • Provider credentials are encrypted at rest with AES-256-GCM and are never passed to the AI client.
  • OAuth refresh tokens and client secrets are stored only as SHA-256 hashes, and access tokens expire after an hour.
  • Every record is scoped to a tenant, and every request is authorized against the tenant that owns the connection.
  • Unauthenticated requests to the MCP endpoint are rate limited per IP address.
  • On Scale and Enterprise, the account owner can enable an account-level IP allowlist of up to 50 entries, IPv4 and IPv6, including CIDR ranges.
  • Telemetry records method, tool name, outcome, latency, size, client and a hashed network address. The contents of provider responses are not stored.
  • Application hosting and the database run in the United States with Hostinger. Every subprocessor is named, with what it handles and where.
  • A Data Processing Agreement under GDPR Article 28 is published, alongside a current subprocessor list.

Established in Australia, used wherever your tools are

PopMCP is operated from Australia, and the customer agreement is governed by Australian law. The published terms, privacy policy, security overview, data processing agreement, and current subprocessor list are all in the legal center rather than available on request.

No provider in the catalog is held back as an upsell. Plans set how many connections and workspaces you run, not which of the 100+ providers you can choose from.

Start with two connections and no card

Every account opens with 14 days on Studio, no card required. After that the free plan runs indefinitely with one workspace, two connections and read-only tools.

Connect your first MCP