
Patchloom
Summary
Structured, format-preserving file editor for agents: JSON/YAML/TOML, markdown, and AST-aware code edits, transactional across files.
Features
- Parser-backed JSON/YAML/TOML edits that preserve comments and formatting
- AST-aware rename/replace/list-symbols across ~20 languages
- Markdown table, bullet, and section editing
- Multi-file batch/transaction edits with rollback
Installation
Set up this MCP server in your favorite AI agent — copy a ready-made configuration below.
Description
Patchloom targets a specific pain point for coding agents: naive text replacement on structured files, whether config, markdown, or source code, is error-prone and burns tool-call round-trips. Instead of shell-quoted sed or patch commands, it gives agents typed, structured operations: doc commands that edit JSON/YAML/TOML while preserving comments and formatting, md commands for editing markdown tables, bullet lists, and sections without rewriting the whole file, ast commands for AST-aware rename, replace, and list-symbols across roughly twenty programming languages, plus generic search/replace/patch/create/delete/read and a tidy command for whitespace normalization.
A batch and transaction mode lets multiple edits across multiple files execute as one transaction with rollback if any step fails, which is the kind of atomicity agents need when refactoring across a codebase. It ships as both a CLI and an MCP server, so an agent can call these operations as structured JSON tool calls rather than generating shell commands the host has to parse and quote correctly. Distribution is unusually broad for a young project, spanning npm, Homebrew, Scoop, Chocolatey, crates.io, an official MCP Registry listing, and VS Code and Open VSX extensions, backed by well over a thousand commits and thousands of tests. This is aimed at coding-agent builders and power users of Claude Code, Cursor, and VS Code who want safer, cheaper structured edits than raw find and replace.
Related MCP Servers
mcp-remote
by Glen Maddern
A proxy that lets local-only MCP clients connect to remote servers, handling OAuth and transport differences.
Blender
by Siddharth Ahuja
Control Blender from an AI client — build scenes, apply materials, and run Python against the live session.
E2B
by E2B
Run AI-generated code safely in an isolated cloud sandbox, with results returned as data rather than text.
ElevenLabs
by ElevenLabs
ElevenLabs' official MCP server — generate speech, clone voices, transcribe audio, and build voice agents.