Sidecar FAQs

Question 1

Can Sidecar manage the output size for LLM interactions?

Accepted Answer

Yes, Sidecar provides configurable output control. The `output` tool accepts a `maxBytes` parameter per request, and there's a global `SIDECAR_MAX_OUTPUT_SIZE` environment variable. This caps the total bytes of stdout/stderr returned to the LLM, preventing large outputs from overwhelming the agent and ensuring efficient context management.

Question 2

How can I install Sidecar into my coding agent?

Accepted Answer

The quickest way is using `npx add-mcp mcp-sidecar`, which can detect and install it for supported agents (Claude Code, Cursor, VS Code, Gemini CLI, etc.). Manual configuration instructions are also provided for specific agents or generic MCP setups, and you can install it from source via `go install` or download pre-built binaries.

Question 3

How does Sidecar ensure command security?

Accepted Answer

Sidecar operates in a 'secure mode' when `SIDECAR_ALLOWED_EXECUTABLES` is configured. This mode enforces an executable allowlist, rejects metacharacters in commands, allows definition of blocked regex patterns, and provides optional audit logging for all start/stop/blocked events, significantly reducing security risks.

Question 4

What problem does Sidecar solve for AI coding agents?

Accepted Answer

AI coding agents typically lack the ability to run long-lived processes while maintaining interaction. Sidecar solves this by allowing agents to start, manage, and interact with processes like API servers, build watches, or databases over time, enabling complex, stateful development workflows.

Question 5

What are Sidecar's core features and benefits?

Accepted Answer

Sidecar offers cross-platform compatibility (Windows, Linux, macOS), zero runtime dependencies as a single Go binary, reliable cleanup of all child processes, configurable output control for LLM responses, and robust command security with optional executable allowlists and audit logging.

Sidecar

Sidecar

主な機能

ユースケース

主な機能

ユースケース