Releases·Tool Release
Built an MCP tool that gives Claude get_symbol() instead of reading whole files — 97% cheaper, benchmarked against grep/Serena/jcodemunch
A developer released Scalpel, an MCP tool for Claude Code that replaces whole-file reads with a get_symbol(name) call returning a definition plus call sites. It benchmarked 97% fewer tokens and beat grep, Serena, and jcodemunch on cost, though it did not speed bug fixes versus a grep-based agent.
4 engagement1 sourceTue, Sep 15, 2026, 04:21 PM
AI Overview
- amritessh's scalpel MCP tool returns get_symbol(name) with the definition plus call sites, cutting Claude Code token usage 97% versus whole-file reads
- scalpel beat grep, Serena, and jcodemunch on token cost under real API billing
- scalpel tied a plain grep-based agent on fix speed across 57 real bugs
- scalpel installs via npm install -g u/amritessh/scalpel; repo at github.com/amritessh/scalpel-mcp
Related
Social·Community
Developer shares two portable changes that made Claude Code work for .NET
A developer building an open-source kit to improve Claude Code for .NET found that most of its 47 skills, 10 subagents, and 16 slash commands were wasted effort. Two portable changes—likely involving better code search than grep—did almost all the work, and the lessons apply to any language.
A developer building an open-source kit to improve Claude Code for .NET found that most of its 47 skills, 10 subagents, and 16 slash commands were wasted effort. Two portable changes—likely involving better code search than grep—did almost all the work, and the lessons apply to any language.
3 engagement1 sourceredditWed, Jul 29, 2026, 11:59 AM
Releases·Tool Release
code-context MCP server gives Claude Code semantic code search
A developer built code-context, an MCP server that indexes a local repository and provides semantic code search to Claude Code, replacing grep-based file crawling. The tool allows Claude to find relevant code by meaning rather than exact keyword matching, reducing wasted file reads and missed context.
A developer built code-context, an MCP server that indexes a local repository and provides semantic code search to Claude Code, replacing grep-based file crawling. The tool allows Claude to find relevant code by meaning rather than exact keyword matching, reducing wasted file reads and missed context.
46 engagement1 sourceredditThu, Jul 16, 2026, 02:40 PM
Releases·Tool Release
Developer builds SymbolPeek MCP server for symbol-level AI code access
A developer created SymbolPeek, an open-source MCP server that gives AI coding agents symbol-level access to TypeScript/JavaScript codebases instead of reading entire files. It uses the official TypeScript language server to resolve symbols, aiming to reduce wasteful file reads when answering targeted questions.
A developer created SymbolPeek, an open-source MCP server that gives AI coding agents symbol-level access to TypeScript/JavaScript codebases instead of reading entire files. It uses the official TypeScript language server to resolve symbols, aiming to reduce wasteful file reads when answering targeted questions.
1 engagement1 sourceredditThu, Jul 23, 2026, 08:29 AM
Products·launched
AST-retrieval MCP for coding agents cuts token use 93%
An open-source MCP server that gives coding agents exact AST slices via tree-sitter instead of whole files, reducing token consumption. Built and tested while developing a Roblox game with Claude Code, it saved 2.1M tokens (93%) over the project. Aimed at developers using LLM coding assistants to lower costs and improve efficiency.
An open-source MCP server that gives coding agents exact AST slices via tree-sitter instead of whole files, reducing token consumption. Built and tested while developing a Roblox game with Claude Code, it saved 2.1M tokens (93%) over the project. Aimed at developers using LLM coding assistants to lower costs and improve efficiency.
11 engagement2 sourcesredditThu, Aug 13, 2026, 07:57 PM
Releases·Tool Release
Developer A/B tests codebase-memory-mcp against grep on production monorepo
A developer ran a controlled A/B test comparing codebase-memory-mcp, a knowledge-graph indexing tool for codebases, against plain grep on a production TypeScript monorepo using Claude Code subagents. The test aimed to verify claims of token savings and sub-millisecond queries. Results are being shared as a community reaction to the tool's marketing.
A developer ran a controlled A/B test comparing codebase-memory-mcp, a knowledge-graph indexing tool for codebases, against plain grep on a production TypeScript monorepo using Claude Code subagents. The test aimed to verify claims of token savings and sub-millisecond queries. Results are being shared as a community reaction to the tool's marketing.
19 engagement1 sourceredditMon, Aug 3, 2026, 08:24 AM