Every tool your AI uses,
approved and logged
One gateway for MCP servers, connectors, and skills, used by xpander agents, Claude, ChatGPT, Cursor, and any MCP client. You decide who gets which tools, down to a single action, and every call is recorded.
One URL
for every AI client
Any client that supports remote MCP servers connects to one xpander URL.
Desktop and code clients
Claude Code, Codex, Claude Desktop, Cursor, and VS Code connect to the same URL.
Cloud clients
Claude on the web and ChatGPT reach the gateway from their own cloud, so they need a URL that is reachable from the internet.
xpander agents
Agents in the Agent Runtime use the same gateway, the same tools, and the same rules.
Air-gapped? Use desktop clients and xpander agents.
What you control
from one gateway
Decide who gets which tools, down to a single action, and keep a record of every call.
One catalog
MCP servers, skills, and ready connectors to your systems, registered in one place. Turn any internal API into tools from its OpenAPI spec.
Access down to the action
Give each group only the tools it needs. An MCP server with 70 tools can show 20 to one group and all 70 to another. Hidden tools can’t be called.
Each person’s own permissions
For each tool, choose a shared service account or each person’s own sign-in, so people only reach the data they could reach anyway.
Credentials in the vault
Tokens never reach the AI client or the model.
Approvals
Sensitive actions wait for a person to approve them.
Every call recorded
Who called which tool, with what result, in the audit log.
Already run MCP servers? Keep them
Keep the servers you run today, from GitHub and Jira MCP to internal, OpenAPI, and local stdio servers. xpander puts them behind one gateway, with access rules and logs.
GitHub MCP
Jira MCP
Internal MCP servers
OpenAPI tools
Local stdio servers
One URL for every client
Access down to the action
Credentials in the vault
Approvals
Every call logged
Skills registry
Skills, clients and servers,
governed the same way
Share approved skills, connect each person’s AI clients, and run MCP servers local or remote, all behind the same access rules and logs.
Skills registry
Share approved skills across agents and teams from one place, and keep private skills visible only to the people who should see them.
Approval for skills agents build
Admins can require approval before a skill an agent built joins the library.
Coding tools in one step
Claude Code, Codex, and OpenCode connect to the AI Gateway with one copied command per tool.
One URL for every client
Claude, ChatGPT, Cursor, and VS Code sign in once at one MCP URL.
Local or remote MCP servers
Run your own MCP servers next to the agent, inside its sandbox, or keep them remote behind the gateway. Either way, access rules and logs apply.
Every tool call,
checked against your rules
Each call is allowed, sent for approval, or blocked by the policy for that person or agent, and every decision is recorded.
Questions we hear from IT leaders
No. Your existing servers go behind the gateway.
Yes. For each tool you choose a shared service account or each person’s own sign-in.
Yes, except for clients that run in a vendor’s cloud.
Same platform
What comes next
The rest of the platform
Add the AI Gateway for models, then the Agent Runtime and AI Workspace, on the same platform.
