01 · providers & modelsConnect the brains you already use.
Settings centralize Anthropic, OpenAI, Google, Kimi, OpenRouter and Mistral-style providers. Use OAuth/subscription flows where supported, API keys where you prefer explicit billing, then choose the model and context behavior per conversation.
- Provider credentials stay local to your machine.
- Switch models without rebuilding the workspace.
- Use different models for main chat, sub-agents or teams.
02 · toolsDecide what the agent can do.
Every built-in tool is visible: read, glob, grep, edit_file, write_file, bash, web, image, todo, question, clean_context, database, MCP, skills, sub-agents and teams.
- Toggle tools on or off.
- Rewrite descriptions to change behavior.
- Keep risky capabilities disabled until needed.
03 · skillsSave repeatable expertise as Markdown.
Skills are long-form playbooks the agent can load on demand. They live on disk in ~/.claakecode/skills, so your workflow is portable, editable and reviewable.
- Create, edit and delete from the UI.
- Document release processes, design rules or QA flows.
- No plugin store, SDK or approval queue.
04 · MCP serversAttach external capabilities safely.
The Model Context Protocol panel lets you add workspace-scoped servers without spelunking through JSON. Once configured, MCP tools become part of the same visible agent surface.
- Add command, args, env and working directory.
- Enable only the servers relevant to the repo.
- Hot-reload so the agent can pick them up quickly.
05 · databasesLet the agent inspect only approved data.
Add Postgres, MySQL, SQLite, MSSQL or Supabase REST sources from Settings. The agent can list sources, describe schemas and query data only when that source is enabled.
- Read-only mode and row limits by source.
- Destructive operations require confirmation.
- Import candidates from environment files, then approve manually.
06 · embeddings & contextControl memory, retrieval and cleanup.
Embedding settings and context controls help long sessions stay useful. The agent can clean noisy tool results, compact conversations, and keep important AGENTS.md and DESIGN.md instructions in view.
- Tune retrieval behavior for your workspace.
- Use compaction when conversations get long.
- Keep prompts auditable instead of invisible.
07 · prodWire deployment providers without leaving the IDE.
The Prod panel groups deployment-provider credentials, CLI checks and project actions for release work where the agent needs context but you still control authentication and destructive steps.
- Check provider CLI and auth status.
- Store tokens intentionally, not in prompts.
- Keep deployment actions visible before execution.
08 · safety defaultsMake autonomy reversible.
Rollback checkpoints, confirmation prompts, todo boards and explicit questions let the agent move fast without pretending it owns the repo.
- Restore files after a bad branch of work.
- Ask before ambiguous or destructive operations.
- Give teams and sub-agents bounded task boards.