Claude Code

Claude Code is the most fully supported CLI in Open Free Max — exact session resume, MCP server injection, project memory as a tool, and exact token accounting.

Claude Code is the reference implementation for Open Free Max: it is the CLI with the richest command-line surface, so it is the one where OFM can do the most.

What OFM does with it

  • Exact session resume. OFM assigns the session id at launch (--session-id <uuid>) and reopens the very same conversation later (--resume <uuid>). A tab always comes back to the conversation it belonged to, even after a crash or a reboot.
  • MCP injection with zero side effects. Custom MCP servers, the OFM memory tool and an Obsidian vault are all passed as --mcp-config files. Your ~/.claude.json is never touched.
  • Exact token accounting. Claude Code writes real input/output/cache token counts to its transcript, so the Consumption view reports measured numbers, not estimates.
  • History ingestion. Past conversations are ingested into project memory, and a search result can reopen the original conversation in a live tab.
  • Distillation. The “distill” feature that turns raw history into durable facts runs on a Claude session — see Project knowledge.

Launch flags

Set once per provider in Agent settings, and applied at the start of every Claude Code session:

Setting Effect
Model Alias or full name (opus, sonnet, claude-opus-4-8). Empty = account default
Reasoning effort Effort level, model dependent
Permission mode default, plan, acceptEdits, auto, dontAsk, bypassPermissions
Skip all permissions Equivalent to --permission-mode bypassPermissions
Additional folders Extra directories the agent may read and edit
Allowed / disallowed tools Auto-approve or deny rules, one per line (Bash(git *), Bash(rm *))
MCP config Extra MCP configuration files to load
Append to system prompt Text appended without replacing the default prompt
IDE connection, Verbose Connect to a detected IDE; turn-by-turn output
Additional arguments, Environment variables Escape hatches for anything unlisted

Permission modes and Mission Control

Permission mode decides who the CLI asks. Mission Control decides who answers. They compose:

  • default + Mission Control rules — the recommended setup. Claude asks; OFM auto-answers the confirmations you have already approved and routes the rest to the queue.
  • plan — read-only exploration. Nothing to approve.
  • bypassPermissions (or “Skip all permissions”) — Claude stops asking entirely. Mission Control then has nothing to intercept: no queue entry, no rule, no audit row for actions that were never announced. Reserve it for a trusted folder.

Billing safety

OFM’s hard rule is that agents run on your subscription, in interactive mode, never through a metered API. Two things can break that on your side:

  1. ANTHROPIC_API_KEY in the environment. Claude Code prefers API billing when that key is present. Check your shell profile and the Environment variables field in Agent settings.
  2. Headless usage outside OFM. OFM itself never passes -p, --print or SDK flags.

Details in Subscription vs API.

Installation

claude is a native binary — no Node.js required. If the tab reports it is missing, use the installation assistant, then reopen the terminal so the PATH reloads. Under WSL, install it inside the distribution: a Windows-side install is invisible to a session running in Ubuntu.

Frequently asked

Does Open Free Max use my Claude subscription or the Anthropic API?
Your subscription. OFM launches the interactive claude binary in a pseudo-terminal and never uses headless mode, -p, or the SDK, so nothing is billed to an API meter.
Why should I not set ANTHROPIC_API_KEY?
If that variable is present in the session environment, Claude Code switches to API billing. OFM never sets it, but a variable inherited from your shell or added in Agent settings will.

Last updated