GitHub - mksglu/claude-context-mode: Stop losing context to large outputs.

Context Mode

The other half of the context problem.

Every MCP tool call in Claude Code dumps raw data into your 200K context window. A Playwright snapshot costs 56 KB. Twenty GitHub issues cost 59 KB. One access log — 45 KB. After 30 minutes, 40% of your context is gone.

Inspired by Cloudflare's Code Mode — which compresses tool definitions from millions of tokens into ~1,000 — we asked: what about the other direction?

Context Mode is an MCP server that sits between Claude Code and these outputs. 315 KB becomes 5.4 KB. 98% reduction.

Install

/plugin marketplace add mksglu/claude-context-mode
/plugin install context-mode@claude-context-mode

Restart Claude Code. Done. This installs the MCP server + a PreToolUse hook that automatically routes tool outputs through the sandbox + slash commands for diagnostics and upgrades.

claude mcp add context-mode -- npx -y context-mode

claude --plugin-dir ./path/to/context-mode

The Problem

MCP has become the standard way for AI agents to use external tools. But there is a tension at its core: every tool interaction fills the context window from both sides — definitions on the way in, raw output on the way out.

With 81+ tools active, 143K tokens (72%) get consumed before your first message. And then the tools start returning data. A single Playwright snapshot burns 56 KB. A gh issue list dumps 59 KB. Run a test suite, read a log file, fetch documentation — each response eats into what remains.

Code Mode showed that tool definitions can be compressed by 99.9%. Context Mode applies the same principle to tool outputs — processing them in sandboxes so only summaries reach the model.

Tools

How the Sandbox Works

Each execute call spawns an isolated subprocess with its own process boundary. Scripts can't access each other's memory or state. The subprocess runs your code, captures stdout, and only that stdout enters the conversation context. The raw data — log files, API responses, snapshots — never leaves the sandbox.

Ten language runtimes are available: JavaScript, TypeScript, Python, Shell, Ruby, Go, Rust, PHP, Perl, R. Bun is auto-detected for 3-5x faster JS/TS execution.

Authenticated CLIs work through credential passthrough — gh, aws, gcloud, kubectl, docker inherit environment variables and config paths without exposing them to the conversation.

When output exceeds 5 KB and an intent is provided, Context Mode switches to intent-driven filtering: it indexes the full output into the knowledge base, searches for sections matching your intent, and returns only the relevant matches with a vocabulary of searchable terms for follow-up queries.

How the Knowledge Base Works

The index tool chunks markdown content by headings while keeping code blocks intact, then stores them in a SQLite FTS5 (Full-Text Search 5) virtual table. Search uses BM25 ranking — a probabilistic relevance algorithm that scores documents based on term frequency, inverse document frequency, and document length normalization. Porter stemming is applied at index time so "running", "runs", and "ran" match the same stem.

When you call search, it returns relevant content snippets focused around matching query terms — not full documents, not approximations, the actual indexed content with smart extraction around what you're looking for. fetch_and_index extends this to URLs: fetch, convert HTML to markdown, chunk, index. The raw page never enters context.

Fuzzy Search

Search uses a three-layer fallback to handle typos, partial terms, and substring matches:

• Layer 1 — Porter stemming: Standard FTS5 MATCH with porter tokenizer. "caching" matches "cached", "caches", "cach".
• Layer 2 — Trigram substring: FTS5 trigram tokenizer matches partial strings. "useEff" finds "useEffect", "authenticat" finds "authentication".
• Layer 3 — Fuzzy correction: Levenshtein distance corrects typos before re-searching. "kuberntes" → "kubernetes", "autentication" → "authentication".

The searchWithFallback method cascades through all three layers and annotates results with matchLayer so you know which layer resolved the query.

Smart Snippets

Search results use intelligent extraction instead of truncation. Instead of returning the first N characters (which might miss the important part), Context Mode finds where your query terms appear in the content and returns windows around those matches. If your query is "authentication JWT token", you get the paragraphs where those terms actually appear — not an arbitrary prefix.

Progressive Search Throttling

The search tool includes progressive throttling to prevent context flooding from excessive individual calls:

• Calls 1-3: Normal results (2 per query)
• Calls 4-8: Reduced results (1 per query) + warning
• Calls 9+: Blocked — redirects to batch_execute

This encourages batching queries via search(queries: ["q1", "q2", "q3"]) or batch_execute instead of making dozens of individual calls.

Session Stats

The stats tool tracks context consumption in real-time. Network I/O inside the sandbox is automatically tracked for JS/TS executions.

Without context-mode, 9.6MB of raw tool output would flood your context window. Instead, 9.6MB (99%) stayed in sandbox — saving ~2,457,600 tokens of context space.

Subagent Routing

When installed as a plugin, Context Mode includes a PreToolUse hook that automatically injects routing instructions into subagent (Task tool) prompts. Subagents learn to use batch_execute as their primary tool and search(queries: [...]) for follow-ups — without any manual configuration.

Bash subagents are automatically upgraded to general-purpose so they can access MCP tools. Without this, a subagent_type: "Bash" agent only has the Bash tool — it can't call batch_execute or search, and all raw output floods context.

The Numbers

Measured across real-world scenarios:

Playwright snapshot — 56.2 KB raw → 299 B context (99% saved) GitHub Issues (20) — 58.9 KB raw → 1.1 KB context (98% saved) Access log (500 requests) — 45.1 KB raw → 155 B context (100% saved) Context7 React docs — 5.9 KB raw → 261 B context (96% saved) Analytics CSV (500 rows) — 85.5 KB raw → 222 B context (100% saved) Git log (153 commits) — 11.6 KB raw → 107 B context (99% saved) Test output (30 suites) — 6.0 KB raw → 337 B context (95% saved) Repo research (subagent) — 986 KB raw → 62 KB context (94% saved, 5 calls vs 37)

Over a full session: 315 KB of raw output becomes 5.4 KB. Session time before slowdown goes from ~30 minutes to ~3 hours. Context remaining after 45 minutes: 99% instead of 60%.

Full benchmark data with 21 scenarios →

Try It

These prompts work out of the box. Run /context-mode:stats after each to see the savings.

Deep repo research — 5 calls, 62 KB context (raw: 986 KB, 94% saved)

Research https://github.com/modelcontextprotocol/servers — architecture, tech stack,
top contributors, open issues, and recent activity. Then run /context-mode:stats.

Git history analysis — 1 call, 5.6 KB context

Clone https://github.com/facebook/react and analyze the last 500 commits:
top contributors, commit frequency by month, and most changed files.
Then run /context-mode:stats.

Web scraping — 1 call, 3.2 KB context

Fetch the Hacker News front page, extract all posts with titles, scores,
and domains. Group by domain. Then run /context-mode:stats.

Large JSON API — 7.5 MB raw → 0.9 KB context (99% saved)

Create a local server that returns a 7.5 MB JSON with 20,000 records and a secret
hidden at index 13000. Fetch the endpoint, find the hidden record, and show me
exactly what's in it. Then run /context-mode:stats.

Documentation search — 2 calls, 1.8 KB context

Fetch the React useEffect docs, index them, and find the cleanup pattern
with code examples. Then run /context-mode:stats.

Requirements

• Node.js 18+
• Claude Code with MCP support
• Optional: Bun (auto-detected, 3-5x faster JS/TS)

Development

git clone https://github.com/mksglu/claude-context-mode.git
cd claude-context-mode && npm install
npm test              # run tests
npm run test:all      # full suite

License

MIT