Summarizer

MCP server that reduces Claude Code context consumption by 98%

87 comments · 4,680 words

Complete Created: Mar 1, 06:24 AM (00:03:29)

Models: Claude Opus 4.5 (analyze) · Gemini 3 Flash (tag) · Gemini 3 Flash (summarize)

Article URL: https://mksg.lu/blog/context-mode (757 words)

Article Summary

Comment Summary

Topics

• Agentic Context Management (21 comments) (Ideas about models managing their own context, pruning irrelevant information, backtracking failed attempts, and treating context like git branches with cherry-picking and rebasing)
• Prompt Cache Economics (12 comments) (Debate about whether context compression breaks prompt caching, with concerns that verbose but cached context might be cheaper than compressed context that invalidates cache)
• Quality vs Token Savings (12 comments) (Questions about whether compressed context produces equivalent output quality, noting extended sessions only matter if reasoning quality holds)
• Hybrid Retrieval Methods (11 comments) (Discussion of combining BM25 with vector search using Model2Vec embeddings, sqlite-vec, and Reciprocal Rank Fusion for better handling of mixed structured and natural language data in tool outputs)
• Backtracking and Pruning (11 comments) (Ideas for automatically detecting retry patterns and pruning failed attempts once correct solution is found, treating context as editable rather than append-only)
• MCP Response Interception Limitations (10 comments) (Clarification that context-mode cannot intercept MCP tool responses because there's no PostToolUse hook in Claude Code; only built-in tools and CLI wrappers benefit from compression)
• Subagent Architecture Benefits (9 comments) (Discussion of spawning subprocesses for work-oriented calls that don't pollute parent context, returning only summarized results to main thread)
• Tool Count Management (5 comments) (Discussion of whether 80+ tools in context is the real problem, suggesting sub-agents for areas of focus rather than compressing everything)
• CLI vs MCP Tradeoffs (4 comments) (Suggestions to use CLI tools like GitHub CLI instead of MCPs for fraction of token cost, with discussion of when each approach is appropriate)
• Extraction Script Reliability (4 comments) (Concerns that compressing git commits to 107 bytes requires LLM to write perfect extraction scripts upfront, risking information loss when scripts are wrong)
• Hook Aggressiveness Concerns (4 comments) (Criticism that blocking all curl/wget for 56KB snapshots is excessive when many API calls return minimal data; author acknowledged and removed)
• Tool Definition Compression (3 comments) (Reference to Cloudflare's Code Mode approach for compressing tool definitions on input side, complementing context-mode's output side compression)
• Structured Data Challenges (3 comments) (Observations that pure BM25 underperforms on tool outputs mixing JSON, tables, config with natural language, requiring hybrid approaches)
• Cross-Platform Compatibility (3 comments) (Questions about support for Codex, Zed Agent, and other platforms beyond Claude Code, noting implementation should be agent-independent)
• Skills vs MCP Debate (3 comments) (Some users suggest using skills and CLI instead of injecting MCP into context, questioning whether skills running in subagents save context)
• Incremental Indexing Performance (2 comments) (Discussion of hashing content for incremental re-embedding of changed chunks only, achieving 10-second updates versus 4-minute full reindexes)
• Context Window Visibility Tools (2 comments) (User built claude-trace CLI to parse usage logs and break down token consumption by session, tool, project, providing measurement before optimization)
• Early Internet Parallels (2 comments) (Observation that current coding agent optimization feels like late 1990s HTML/SQL era, with experienced engineers quickly spotting bottlenecks)
• Dataframe Approach for Logs (1 comment) (Alternative approach creating in-memory parquet dataframes with token-optimized summary views for database and log system responses)

Raw Files

Execution Log

[2026-03-01T14:24:26.469Z] Starting step: fetch_pages (attempt 1)
[2026-03-01T14:24:26.490Z] Fetching HN page: https://news.ycombinator.com/item?id=47193064
[2026-03-01T14:24:26.663Z] Fetched HN page: 145708 bytes
[2026-03-01T14:24:26.850Z] Extracted title: MCP server that reduces Claude Code context consumption by 98%
[2026-03-01T14:24:26.867Z] Extracted linked URL: https://mksg.lu/blog/context-mode
[2026-03-01T14:24:26.883Z] Fetching linked article: https://mksg.lu/blog/context-mode
[2026-03-01T14:24:27.246Z] Fetched linked article: 38129 bytes
[2026-03-01T14:24:27.388Z] Completed step: fetch_pages in 904ms
[2026-03-01T14:24:27.498Z] Starting step: extract_text (attempt 1)
[2026-03-01T14:24:27.611Z] Extracted HN text: 34746 chars
[2026-03-01T14:24:27.741Z] Extracted 87 comments
[2026-03-01T14:24:27.827Z] Extracted linked article text: 4659 chars, 757 words
[2026-03-01T14:24:27.976Z] Comment word count: 4680
[2026-03-01T14:24:28.088Z] Completed step: extract_text in 573ms
[2026-03-01T14:24:28.200Z] Starting step: analyze_content (attempt 1)
[2026-03-01T14:24:28.288Z] Calling claude-opus-4-5-20251101 (article: 4659 chars, 87 comments)
[2026-03-01T14:24:55.935Z] Analysis complete: 19 topics, 8142 input tokens, 970 output tokens
[2026-03-01T14:24:55.965Z] Completed step: analyze_content in 27750ms
[2026-03-01T14:24:56.075Z] Starting step: tag_comments (attempt 1)
[2026-03-01T14:24:56.100Z] Tagging 87 comments with 19 topics (batch size: 50)
[2026-03-01T14:24:56.116Z] Processing batch 1/2 (50 comments)
[2026-03-01T14:25:40.305Z] Batch 1 complete: 78 tags assigned
[2026-03-01T14:25:40.320Z] Processing batch 2/2 (37 comments)
[2026-03-01T14:26:02.058Z] Batch 2 complete: 51 tags assigned
[2026-03-01T14:26:02.072Z] Tagging complete: 129 total tags, 10640 input tokens, 2005 output tokens
[2026-03-01T14:26:02.087Z] Completed step: tag_comments in 65997ms
[2026-03-01T14:26:02.204Z] Starting step: summarize_topics (attempt 1)
[2026-03-01T14:26:02.224Z] Summarizing 19 topics
[2026-03-01T14:26:02.264Z] Summarizing topic 1/19: "Hybrid Retrieval Methods # Discussion of combining BM25 with vector search using Model2Vec embeddings, sqlite-vec, and Reciprocal Rank Fusion for better handling of mixed structured and natural language data in tool outputs" (11 comments)
[2026-03-01T14:26:08.716Z] Topic 1 summarized (1378 in, 146 out)
[2026-03-01T14:26:08.742Z] Summarizing topic 2/19: "MCP Response Interception Limitations # Clarification that context-mode cannot intercept MCP tool responses because there's no PostToolUse hook in Claude Code; only built-in tools and CLI wrappers benefit from compression" (10 comments)
[2026-03-01T14:26:16.041Z] Topic 2 summarized (1199 in, 150 out)
[2026-03-01T14:26:16.064Z] Summarizing topic 3/19: "Prompt Cache Economics # Debate about whether context compression breaks prompt caching, with concerns that verbose but cached context might be cheaper than compressed context that invalidates cache" (12 comments)
[2026-03-01T14:26:24.186Z] Topic 3 summarized (1179 in, 135 out)
[2026-03-01T14:26:24.208Z] Summarizing topic 4/19: "Agentic Context Management # Ideas about models managing their own context, pruning irrelevant information, backtracking failed attempts, and treating context like git branches with cherry-picking and rebasing" (21 comments)
[2026-03-01T14:26:31.354Z] Topic 4 summarized (1753 in, 144 out)
[2026-03-01T14:26:31.376Z] Summarizing topic 5/19: "Subagent Architecture Benefits # Discussion of spawning subprocesses for work-oriented calls that don't pollute parent context, returning only summarized results to main thread" (9 comments)
[2026-03-01T14:26:38.440Z] Topic 5 summarized (1044 in, 154 out)
[2026-03-01T14:26:38.463Z] Summarizing topic 6/19: "CLI vs MCP Tradeoffs # Suggestions to use CLI tools like GitHub CLI instead of MCPs for fraction of token cost, with discussion of when each approach is appropriate" (4 comments)
[2026-03-01T14:26:43.531Z] Topic 6 summarized (365 in, 132 out)
[2026-03-01T14:26:43.556Z] Summarizing topic 7/19: "Tool Definition Compression # Reference to Cloudflare's Code Mode approach for compressing tool definitions on input side, complementing context-mode's output side compression" (3 comments)
[2026-03-01T14:26:50.211Z] Topic 7 summarized (433 in, 143 out)
[2026-03-01T14:26:50.233Z] Summarizing topic 8/19: "Incremental Indexing Performance # Discussion of hashing content for incremental re-embedding of changed chunks only, achieving 10-second updates versus 4-minute full reindexes" (2 comments)
[2026-03-01T14:26:55.464Z] Topic 8 summarized (675 in, 103 out)
[2026-03-01T14:26:55.486Z] Summarizing topic 9/19: "Extraction Script Reliability # Concerns that compressing git commits to 107 bytes requires LLM to write perfect extraction scripts upfront, risking information loss when scripts are wrong" (4 comments)
[2026-03-01T14:27:01.106Z] Topic 9 summarized (477 in, 133 out)
[2026-03-01T14:27:01.128Z] Summarizing topic 10/19: "Context Window Visibility Tools # User built claude-trace CLI to parse usage logs and break down token consumption by session, tool, project, providing measurement before optimization" (2 comments)
[2026-03-01T14:27:05.653Z] Topic 10 summarized (260 in, 116 out)
[2026-03-01T14:27:05.674Z] Summarizing topic 11/19: "Structured Data Challenges # Observations that pure BM25 underperforms on tool outputs mixing JSON, tables, config with natural language, requiring hybrid approaches" (3 comments)
[2026-03-01T14:27:10.802Z] Topic 11 summarized (478 in, 114 out)
[2026-03-01T14:27:10.824Z] Summarizing topic 12/19: "Quality vs Token Savings # Questions about whether compressed context produces equivalent output quality, noting extended sessions only matter if reasoning quality holds" (12 comments)
[2026-03-01T14:27:18.558Z] Topic 12 summarized (1031 in, 133 out)
[2026-03-01T14:27:18.579Z] Summarizing topic 13/19: "Hook Aggressiveness Concerns # Criticism that blocking all curl/wget for 56KB snapshots is excessive when many API calls return minimal data; author acknowledged and removed" (4 comments)
[2026-03-01T14:27:23.797Z] Topic 13 summarized (436 in, 117 out)
[2026-03-01T14:27:23.820Z] Summarizing topic 14/19: "Dataframe Approach for Logs # Alternative approach creating in-memory parquet dataframes with token-optimized summary views for database and log system responses" (1 comments)
[2026-03-01T14:27:29.237Z] Topic 14 summarized (313 in, 116 out)
[2026-03-01T14:27:29.259Z] Summarizing topic 15/19: "Backtracking and Pruning # Ideas for automatically detecting retry patterns and pruning failed attempts once correct solution is found, treating context as editable rather than append-only" (11 comments)
[2026-03-01T14:27:35.163Z] Topic 15 summarized (1326 in, 145 out)
[2026-03-01T14:27:35.187Z] Summarizing topic 16/19: "Cross-Platform Compatibility # Questions about support for Codex, Zed Agent, and other platforms beyond Claude Code, noting implementation should be agent-independent" (3 comments)
[2026-03-01T14:27:39.904Z] Topic 16 summarized (179 in, 79 out)
[2026-03-01T14:27:39.924Z] Summarizing topic 17/19: "Skills vs MCP Debate # Some users suggest using skills and CLI instead of injecting MCP into context, questioning whether skills running in subagents save context" (3 comments)
[2026-03-01T14:27:44.292Z] Topic 17 summarized (208 in, 106 out)
[2026-03-01T14:27:44.318Z] Summarizing topic 18/19: "Tool Count Management # Discussion of whether 80+ tools in context is the real problem, suggesting sub-agents for areas of focus rather than compressing everything" (5 comments)
[2026-03-01T14:27:49.115Z] Topic 18 summarized (371 in, 115 out)
[2026-03-01T14:27:49.137Z] Summarizing topic 19/19: "Early Internet Parallels # Observation that current coding agent optimization feels like late 1990s HTML/SQL era, with experienced engineers quickly spotting bottlenecks" (2 comments)
[2026-03-01T14:27:53.488Z] Topic 19 summarized (229 in, 108 out)
[2026-03-01T14:27:53.504Z] Summarization complete: 19 topics, 13334 input tokens, 2389 output tokens
[2026-03-01T14:27:53.518Z] Completed step: summarize_topics in 111299ms
[2026-03-01T14:27:53.557Z] Job completed successfully

LLM Invocations (Total: $0.0901)

← Back to all jobs