Integrations

Your tools, reading your files.

GNO installs into ten clients automatically and works with any other MCP client through manual configuration. Each install pins an absolute runtime path plus the active index, config, data, and model cache, so a desktop app opens the same workspace as your shell without inheriting environment variables it never sees. Read-only by default: the twenty-one write tools, gno_remember among them, require an explicit flag.

The short version

Two ways in: an agent skill or an MCP server.

Pair either connector with the retrieval protocol for supported harnesses. It teaches when to retrieve and how to maintain a second brain or LLM wiki. Skills suit agents that already shell out. MCP exposes structured tools, including Context Capsule build and verification, Knowledge Delta, and the opt-in verified Ask.

# Choose your agent target: claude, codex, opencode, openclaw, hermes
gno skill install --target claude --scope user
gno agents install --target claude
gno agents verify --target claude
# MCP clients: Claude Desktop, Cursor, Zed, Windsurf, Amp, LM Studio, LibreChat
gno mcp install --target cursor
# Slim 7-tool core profile, chosen once per client
gno mcp install --target cursor --tool-profile core
# Inspect exactly what would be written, without writing it
gno mcp install --target cursor --dry-run --json

Anything else

Any MCP client works, it just needs manual configuration.

Raycast and other compatible clients point at the same stdio or Streamable HTTP endpoint. The MCP reference has the exact command, arguments, and workspace values to copy.