Skip to content

Garry's Opinionated OpenClaw/Hermes Agent Brain

TypeScript
2.6 KB
Raw Blame

Harness adapter reference

A supported transport is distinct from a tested vendor session. The new Grok Bot and Muse paths use finite local or thin CLI calls; native MCP activation is not asserted. Query gbrain mcp adapters for dated documentation sources. Token lifetimes below apply to new grants unless explicitly overridden.

Adapter Supported modes Connection mechanism New access-token lifetime Instructions
claude-code local-cli, stdio, http claude-json 30 days Guide
claude-desktop stdio, http Follow client settings 1 hour Guide
codex local-cli, stdio, http codex-toml 30 days Guide
opencode local-cli, stdio, http opencode-json 30 days Guide
openclaw local-cli, thin-cli thin-cli 1 hour Guide
grok-build local-cli, stdio, http Follow client settings 1 hour Guide
grok-bot local-cli, thin-cli thin-cli 1 hour Guide
muse local-cli, thin-cli thin-cli 1 hour Guide
muse-code stdio, http Follow client settings 1 hour Guide
hermes stdio Follow client settings 1 hour Guide
cursor http Follow client settings 1 hour Guide
perplexity http Follow client settings 1 hour Guide
chatgpt http Follow client settings 1 hour Guide
generic http Follow client settings 1 hour Guide

Aliases: grok means Grok Build; claude means Claude Code. The older --agent connection commands retain their defaults. Use the separate grok-bot, muse, and muse-code identifiers for those products.

Managed installers preserve unrelated configuration and use private credential files and ownership receipts. Manual adapters give guidance; they do not write a universal configuration format. Reload behavior and native enablement steps are recorded in the install result.

See hosted access for profiles, private delivery, removal, and verification. Actual harness evidence remains required after server checks pass.