Sipcode

What is Sipcode?

Sipcode is an open-source (MIT) MCP server and PreToolUse hook for Claude Code. It keeps the AI’s context clean over long sessions and exposes context hygiene to Claude as MCP tools.

How to use Sipcode?

Install globally via npm: npm i -g sipcode && sipcode init. After setup, it runs automatically within Claude Code sessions.

Key features of Sipcode

• Caps verbose tool output (grep, npm install, git log, tsc) before it reaches the model.
• Deduplicates same-session re-reads of unchanged files using content-hash checks.
• Ships 15 MCP tools for inspecting context stats: token waste, duplicate reads, drift, cost forecasts.
• Reports ~62% median token savings on a reproducible benchmark.
• Open-source under MIT license.

Use cases of Sipcode

• Maintaining AI reliability during long Claude Code sessions.
• Reducing token consumption by filtering verbose command outputs.
• Avoiding redundant reading of unchanged files across multiple tool calls.
• Monitoring context health mid-session with provided MCP tools.
• Improving code generation quality by keeping the context window clean.

FAQ from Sipcode

What problem does Sipcode solve?

It addresses context window bloat by capping verbose output and deduplicating file reads, keeping the AI reliable over long sessions.

How does Sipcode deduplicate file reads?

It uses content-hash based deduplication and never serves stale bytes to the model.

What MCP tools does Sipcode provide?

It provides 15 MCP tools for inspecting context stats such as token waste, duplicate reads, drift, and cost forecasts.

What are the reported benefits?

A reproducible benchmark shows ~62% median token savings; Anthropic’s research reports a 29% quality lift and around 40% fewer agent errors when context stays clean.

Is Sipcode open source?

Yes, it is released under the MIT license.