Agent access
Use the Infralane dashboard directly, or connect an MCP-compatible agent to selected projects. Both work with the same applications and deployments.
The Model Context Protocol (MCP) gives a client a structured way to discover platform context, prepare deployments and inspect results. Infralane exposes a remote server over Streamable HTTP. It does not run an embedded language model on your behalf.
What an agent can do
- Inspect authorized projects, existing resource pools, repositories and applications.
- Prepare and apply a deployment from GitHub, a Docker image or a validated ZIP upload.
- Inspect an operation, deployment status and deployment history.
- Read finite build and runtime log snapshots.
The interface supports web services, static sites and workers. Source builds currently support Python, Node.js and static runtimes. A Docker image supplies its own runtime.
What stays outside this interface
MCP tools do not purchase resource pools, scale pools, execute terminal commands, restart or roll back applications, delete applications, or manage custom domains and persistent volumes. Do not infer an agent operation from the existence of a dashboard control.
A deployment plan is a configuration preview, not a running preview environment. Applying a change to an existing application can cause downtime.
Before you connect
Create an account, workspace and project, then connect your own server for free management. It becomes the resource pool for agent deployments. For GitHub sources, connect the workspace GitHub integration and grant access to the desired repository. Agent tools use existing resources and do not purchase capacity.
Client configuration and authorization support vary. See Connect an agent for the required capabilities. A general MCP implementation does not guarantee compatibility with every client.