Overview
Parsaa’s agent works by calling tools — to read code, edit files, run builds, and more. You control this on two independent axes:
- Permissions — whether a category of action runs freely, prompts you, or is blocked.
- Enabled tools — which individual tools are available at all.
Capabilities
Every tool belongs to a capability bucket. Permissions are set per capability, not per tool.
| Capability | What it covers |
|---|
| Read & search | Read files, grep, search, symbols, git status/diff |
| Edit code | Write changes to files (create_file, edit_file, rewrite_file, delete_file) |
| Git | Git commits |
| Run commands | Run shell commands (execute_command) |
| Build & run app | Compile, run, test, and control simulators/devices |
| MCP tools | Tools from connected MCP servers |
Permission Modes
Each capability is set to one of three modes:
| Mode | Behavior |
|---|
| Allow | Runs immediately, no prompt |
| Ask | Prompts you before running |
| Deny | Never runs — the tool isn’t even offered to the model |
Defaults are safe: Read & search is Allow (read-only runs freely), and every mutating capability (edit, git, run commands, build) defaults to Ask.
The approval prompt
When a capability set to Ask is invoked, Parsaa pauses and shows an inline approval card with a preview of the exact action (e.g. the shell command or file path). You choose:
| Choice | Effect |
|---|
| Deny | Reject this call |
| Allow for session | Allow this capability for the rest of the chat session |
| Allow once | Allow just this single call |
- Built-in tab — toggle each built-in tool; a
? shows its description. New tools appear automatically.
- MCP tab — tools grouped by server (see MCP Tools).
- Bulk Enable All / Disable All, plus search.
A disabled tool is never offered to the agent, regardless of its capability’s permission mode.
| Capability | Tools |
|---|
| Read & search | file_tree, read_files, grep_project, globs, search_file_names, get_symbol_info, semantic_search, git_status, git_diff, list_schemes, list_simulators, list_devices, show_build_settings, get_app_bundle_id |
| Edit code | create_file, edit_file, rewrite_file, delete_file |
| Git | git_commit |
| Run commands | execute_command |
| Build & run app | compile_app, run_app, test_app, boot_simulator, install_app, launch_app, stop_app |
| MCP tools | any mcp_* tool from a connected server |
Sub-agents get their own tool allowlist and their own per-capability permission matrix, so you can make a sub-agent stricter than the main agent. See Sub-agents. 