context window plugins

msiShariful/claude-token-inspector

Plugin Claude Code

Claude Code plugins by Shariful Islam — token inspection and context optimization tools.

msiShariful/claude-token-inspector

Plugin Claude Code

Inspect and identify bloated context — find large files, redundant instructions, and unnecessary content consuming your token budget.

Open330/context-compress

Plugin Claude Code

MCP server and hook toolkit that compresses responses and keeps explicitly indexed output searchable.

Open330/context-compress

Plugin Claude Code

Compress tool responses and keep explicitly indexed output searchable without flooding the agent context window.

munhq/distil

Plugin Claude Code

Context optimization for LLM agents: measure where a session's tokens actually go, and compress context only where compression pays for the prompt cache it invalidates.

munhq/distil

Plugin Claude Code

Skills and MCP server for context optimization — decide whether compressing an agent's context pays for the prompt cache it invalidates, and compress it when the answer is yes. Bundles the distil MCP server so install wires everything in one step.

ofershap/prompt-compression

Plugin Claude Code

Compress documentation, prompts, and context into minimal tokens for AGENTS.md. Based on Vercel's research showing compressed passive context achieves 100% pass rate vs 53% baseline.

ignatenkofi/memshelf-mcp

Plugin Claude Code

Working-memory shelf for Claude Code: the /shelve skill plus a SessionStart hook that injects the shelf INDEX (recall bootstrap) and a SessionEnd/PreCompact hook that pushes the shelf so episodes survive an ephemeral container.

Rovemark/logica-context

Plugin Claude Code

Protects your AI context window via sandboxed execution and smart indexing.

Cambrionic/winnow

Plugin Claude Code

Winnow — use many MCP servers without context bloat.

Cambrionic/winnow

Plugin Claude Code

Use many MCP servers without context bloat. Winnow hides all your MCP servers behind one gateway exposing just 4 tools (searchtools / loadtool / calltool / runcode), so the model spends a fraction of the context.