8. MCP (Model Context Protocol)

Rayu can connect to MCP servers to expose extra tools and resources to the model. MCP works the same as in the upstream CLI.

Managing servers

# stdio server (a local command)
rayu mcp add my-server -- my-command --flag arg1

# HTTP/SSE server
rayu mcp add --transport http sentry https://mcp.sentry.dev/mcp

# with headers / env
rayu mcp add --transport http corridor https://app.corridor.dev/api/mcp --header "Authorization: Bearer ..."
rayu mcp add -e API_KEY=xxx my-server -- npx my-mcp-server

# list (runs a health check)
rayu mcp list

# show one server
rayu mcp get my-server

# remove
rayu mcp remove my-server

add options include:

Option	Meaning
`-s, --scope <scope>`	`local`, `user`, or `project`
`-t, --transport <t>`	`stdio` (default), `sse`, or `http`
`-e, --env <KEY=value>`	Environment variables for a stdio server
`-H, --header <h>`	Headers for HTTP/SSE servers

You can also manage servers in-session with /mcp.

Where MCP config lives

MCP servers are stored in the global ~/.claude.json (project-scoped entries), which is shared with Claude Code. If RAYU_CONFIG_DIR / CLAUDE_CONFIG_DIR is set, the file lives inside that directory instead.

Project-level servers can also be declared in a .mcp.json file in your repo, or loaded ad hoc with --mcp-config.

Using MCP tools

Once a server is connected (✓ Connected in rayu mcp list), its tools appear to the model namespaced as mcp__<server>__<tool> and can be called during a conversation. Resources are available via the built-in ListMcpResources / ReadMcpResource tools.

OpenAI-compatible providers + MCP

MCP works regardless of provider. Tool schemas are translated to the active provider's format automatically (Anthropic tools for the Anthropic path; OpenAI function tools for OpenAI-compatible providers), so MCP tools are usable with NVIDIA/DeepSeek/etc. as well as Anthropic.

Next: Diagnostics & Privacy →