Skip to content

README.md

Latest commit

 

History

History
92 lines (69 loc) · 3.47 KB

README.md

File metadata and controls

92 lines (69 loc) · 3.47 KB
Raw

SuperBased Observer — Open Interpreter

Local-first token, cost and cache observability for Open Interpreter (interpreter).

Open Interpreter documents no plugin or extension mechanism at all — MCP config is its entire extension surface. So this is a config listing: the exact TOML table to paste.

Requires the observer binary already on your PATH. This is wiring only — the table below declares an MCP server that runs observer; it does not download or bundle it.

npm i -g @superbased/observer     # or: pipx install superbased-observer
observer start                    # run the local daemon

Install

Append this to ~/.openinterpreter/config.toml:

  [mcp_servers.observer]
    args = ["serve"]
    command = "observer"

Then check it with interpreter mcp list, or /mcp inside the TUI.

(The indentation is the TOML writer's, not a typo — leading whitespace before a table header and its keys is legal TOML, and this block is reproduced exactly as observer's own writer emits it. See "Where this block comes from".)

What it wires

Key Value
command + args observer serve — the same launch observer init writes for every other client, resolved from PATH.

Open Interpreter also documents env, enabled, required, enabled_tools, disabled_tools, startup_timeout_sec and tool_timeout_sec on a server table. None is emitted: observer's server takes no environment, starts locally in milliseconds, and pre-disabling one of its own tools on your behalf is not this listing's decision to make.

The observer server exposes observer's project, session, cost and cache queries as tools, reading the local ~/.observer/observer.db. It makes no network calls of its own.

Open Interpreter capture is unaffected either way: observer's watcher reads its rollout JSONL under ~/.openinterpreter/sessions whether or not this table exists.

Where this block comes from

It is not hand-typed. Open Interpreter is a rebadged Codex CLI build and shares Codex's [mcp_servers.<name>] config format, so the text above is the exact table observer's real Codex MCP registrar writes — lifted verbatim out of a throwaway sandbox home and documented against ~/.openinterpreter/config.toml instead of ~/.codex/config.toml. A changed launch argument reaches this page automatically.

The one difference from what observer init --codex writes is the file it goes in: Open Interpreter keeps "product-only config and session state local under ~/.openinterpreter", its own home. observer init does not write that file — there is no Open Interpreter client row in internal/mcp — which is why this is a paste rather than a command.

Double-wiring

observer init writes no Open Interpreter config, so this hand-added table cannot duplicate anything observer wrote. Note that observer init --codex writes the SAME table into ~/.codex/config.toml for the Codex CLI — a different tool reading a different file, not a duplicate.

The automatic detect-and-skip observer init performs for the Claude Code plugin is claude-code-only (internal/claudeplugin); no equivalent probe exists here. Documented, not built.

Links