The Model Context Protocol — MCP for short — is Anthropic’s open protocol for how LLMs read from and write to external systems. A year after it shipped, it is the closest thing the industry has to a standard. Claude Desktop speaks it. Cursor speaks it. Windsurf speaks it. A growing list of internal platforms at large companies speak it.
Sempleo speaks MCP both ways. That is a deliberate architectural choice, and it is one of the reasons the product sits where it does in your stack.
Sempleo as an MCP server. Point any MCP-aware client at your Sempleo workspace and it reads the same five-layer context your own agents do — scoped to the user, authority-respecting, and fully audited. Open Claude Desktop, connect it to your Sempleo workspace, and your company voice, team vocabulary, client history, project state, and personal preferences are all available through the protocol. Write operations — update an entry, propose a new one — land as pending review, never as a silent mutation. The same governance that protects your agent runs protects external clients.
This is the part most “context platforms” miss. They build a beautiful internal surface and lock it behind a proprietary API. The minute a team wants to use a different LLM client — which every team eventually does — the vendor has to write a new integration, and the team loses time they cannot afford.
We do not want to be in the integration business. So Sempleo exposes its surface through the protocol that already exists.
Sempleo as an MCP client. The other direction is at least as important. Most enterprise data does not, and should not, live in Sempleo. It lives in a warehouse, a vertical SaaS, an internal tool, a compliance system. Those systems either already expose an MCP server or will over the next year — and Sempleo is a first-class client of those servers.
Register the MCP servers your team runs and Sempleo calls them live at retrieval time. Circuit breakers and rate limits are built in; every call lands in the audit log. Professional-plan workspaces can register up to three external servers; Enterprise removes the cap entirely.
This is the move that keeps Sempleo from becoming yet another silo. Your data of record stays where it lives. Sempleo references it, does not copy it. Revoke access and the context evaporates. No shadow copy, no slow-motion data exfiltration.
Why bidirectional matters. The usual pattern in this category is vendor-as-silo: you pour data in, you query through their UI, you are locked in at the data layer. MCP breaks that pattern in two ways at once. Because Sempleo is a server, you keep your choice of LLM client. Because Sempleo is a client, you keep your existing systems of record. Two independent degrees of freedom — LLM above and data below — that most of our competitors lock down in one or both directions.
I have seen enough platform-as-prison products to have views on this. The architecture choice above is the one that lets a team bring Sempleo in, get value in the first month, and bring it out in the last month without a migration project. If we do our jobs right, you will not want to bring it out. But you should be able to.
The platform page has a diagram of where Sempleo sits in the stack, and the access page is the way in for founding customers.
