AI agents and MCP

Every service exposes its main operations as MCP tools: pages_deploy, apps_create_app, apps_deploy, env names (never values), monitors, errors and more. An agent connected to them can do exactly what you can, no more.

Through the CLI

claude mcp add xvii -- xvii mcp

Or in a client's JSON configuration:

{ "mcpServers": { "xvii": { "command": "xvii", "args": ["mcp"] } } }

xvii mcp asks home which services the instance runs, gathers their tools and forwards calls with your CLI sign-in. Add --context NAME for another instance.

Over HTTP

home serves every installed service's tools at one endpoint:

claude mcp add --transport http xvii https://xvii.network/mcp \
  --header "Authorization: Bearer $XVII_TOKEN"

Use a token with the narrowest scope that fits (read for an agent that only looks).

The documentation itself

The docs service has two tools, search_docs and read_doc, at https://docs.xvii.network/mcp, so an agent can look things up in the documentation of the version this instance runs. Every page is also plain Markdown (add .md to its address), and https://docs.xvii.network/llms.txt lists them all.

claude mcp add --transport http xvii-docs https://docs.xvii.network/mcp