SkillAgentSearch skills...

DevProjex

Repomix alternative with up to 40% fewer tokens. Turn any codebase or Git repository into safe, token-efficient context for LLM and coding agents like Claude Code, Codex, and Cursor.

Install / Use

claude mcp add Avazbek22 -- npx -y github:Avazbek22/DevProjex

If the server publishes to npm under a different name, use that package instead — check the repo README.

About this skill
🔌

MCP Server

Model Context Protocol server

Quality Score

87/100

Supported Platforms

Claude Code
Claude Desktop
Cursor
OpenAI Codex

Tags

Our assessment of DevProjex

DevProjex scores 87/100 on our quality scale, 485th of 950 AI & Machine Learning skills we index.

Its MCP Server is 30 KB long, well organised into 27 sections with 6 code examples: a thorough specification that gives an agent plenty to work with.

It has 44 GitHub stars, so there is little community track record yet; judge it on its content.

Substance
30/30
Structure
20/20
Description
15/15
Adoption
7/20
Freshness
15/15

Maintenance, license and trust

  • The repository was last updated today, so DevProjex is actively maintained.
  • It is released under the Apache-2.0 license, a permissive license that allows use, modification and commercial use with attribution.
  • Its trust signals score 97/100, with no cautions. These come from repository metadata, not a code audit — read the skill file before letting an agent act on it.

Safety scan

No issues found

Our scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands.

Automated pattern scan on 2026-11-22. It catches known dangerous patterns, not every risk — read a skill before letting an agent act on it.

DevProjex compared with similar skills

All 4 of these similar skills score higher than DevProjex; compare them before choosing.

SkillScoreStarsUpdatedFormat
DevProjex (this skill)by Avazbek228744todayMCP Server
claude-memby thedotmack100119.0ktodayCLAUDE.md
Agent-Reachby Panniantong100114.0ktodayCLAUDE.md
Understand-Anythingby Egonex-AI10093.7ktodayCLAUDE.md
hyperframesby heygen-com10088.2ktodayCLAUDE.md

Frequently asked questions

How do I install DevProjex?
Run claude mcp add Avazbek22 -- npx -y github:Avazbek22/DevProjex. The install tabs above show the steps for each supported agent.
Which AI agents does DevProjex work with?
It is written for Claude Code, Claude Desktop, Cursor and OpenAI Codex, as a MCP Server file. Other agents that read the same format can often use it too.
Is DevProjex safe to use?
Our scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands. It is Apache-2.0-licensed and scores 97/100 on trust signals. Skills are instructions an agent will follow, so read the file before installing it and do not approve commands you do not understand.
Is DevProjex still maintained?
The repository was last updated today, so DevProjex is actively maintained.

DevProjex: MCP / GUI / CLI / TUI all in one

🏆 Selected by the Avalonia UI team for the App Showcase

Downloads npm NuGet Build License Last commit Platforms Glama score

Clean, token-efficient codebase context for LLMs, AI chats, and coding agents. You see exactly what leaves your project, and secrets are masked before it goes.

A Repomix alternative: agents used up to 40% fewer context tokens in head-to-head benchmarks.

DevProjex turns any folder or Git repository into clean context for AI. Paste it into ChatGPT or Claude, or let Claude Code, Codex, and Cursor read the project through its read-only MCP server. A desktop app, a Terminal Workspace, a CLI, and the MCP server share one engine, one set of filters, and one masking pipeline, on Windows, macOS, and Linux.

Tick what you need in a file tree, check the live preview and its token estimate, then copy or export the result as ASCII, Markdown, JSON, or XML, or as a clean copy of the project in a folder or ZIP.

🔒 Read-only and telemetry-free by design. DevProjex does not upload your project contents or collect telemetry.

App demo 🖼️

DevProjex demo: tick files in the desktop app, preview the exact context with its token estimate, mask secrets, compress code, then do the same in the Terminal Workspace

Download 🚀

Microsoft Store: 👉 DevProjex

Latest GitHub release (Windows, macOS, Linux): 👉 github.com/Avazbek22/DevProjex/releases/latest

WinGet (Windows): winget install OlimoffDev.DevProjex

Run without installing

The CLI, the Terminal Workspace, and the MCP server run straight from npm or NuGet, or from a direct release binary:

npx devprojex tree .    # Node.js 20+
dnx devprojex tree .    # .NET SDK 10
./DevProjex tree .      # release binary (DevProjex.exe on Windows)

The npm and NuGet packages contain the CLI, the Terminal Workspace, and the MCP server, but not the desktop app. Requirements per platform, the direct headless archives, and the non-root Docker image are all covered in Docs/Installation.md.

Why DevProjex? 💡

Copying files into a chat window one by one doesn't scale. Packing tools are fast, but you see the result after it's packed, not while you choose. And a coding agent with plain file access reads .env as easily as Program.cs, unless every tool it uses is told to ignore it.

DevProjex works differently: context engineering that is visual, precise, and local-first.

  • You see what leaves your project: file tree, live preview, token estimate, and an agent journal that keeps a receipt of every MCP session.
  • You control what leaves it: Smart Ignore, Git modes, secret and private-data masking, name filters, saved profiles.
  • Your agent gets the same rules: the MCP server is read-only, secret masking there has no off switch, and with Live Context the files you tick are the agent's focus.
  • You get more than text: a real project copy as a folder or ZIP.

DevProjex vs Repomix: the numbers

Token savings measured in blind agent benchmarks: the same tasks and the same model on both sides, each MCP server under a neutral name, and a judge that sees only Answer A and Answer B. In the two 18-task runs of 12 November 2026 (Claude Haiku 4.5 in Claude Code), agents used 40.9% and 28.8% fewer context tokens through DevProjex than through Repomix 1.19.0.

| Measured | DevProjex | Repomix | |---|---:|---:| | Context tokens, 18 tasks, 12 Nov 2026 (DevProjex 059f95af, Repomix 1.19.0) | 8,729,849 | 14,780,675 | | Context tokens, 18 tasks, 12 Nov 2026 (DevProjex 1f90ac4b, Repomix 1.19.0) | 8,868,513 | 12,456,603 | | Required files named, of 84 (same run) | 78 | 66 | | Answers the blind judge preferred, of 18 (same run) | 15 | 2 | | Cost of the 18 sessions (same run) | $2.41 | $3.34 | | Tokens, 11 tasks in 6 languages, 29–30 Sep 2026 (DevProjex 02c0fde7, Repomix 1.18.1) | 973,020 | 1,174,907 | | Answers the blind judge trusted, of 22 (same series) | 17 | 3 | | Files with a planted credential that stay readable, of 7 (11 Sep 2026) | 7 | 1 |

In the second run Claude Code's own file tools (Read, Grep, Glob) were measured too: DevProjex cut token usage by 16.5% against them.

Neither tool leaked any of the seven planted credentials; Repomix drops a file once it finds a secret in it, while DevProjex masks the value and keeps the code around it. Repomix is the stronger choice for broad "how does X work" questions, where packing once and searching the pack is cheaper. The tasks were written by the author, and the full tables are published: the November runs in Docs/Benchmarks, the September series with the agents' and the judge's verdicts verbatim in Docs/Benchmark-History.md.

Use it for

  • AI assistants: clean input for ChatGPT, Claude, DeepSeek, Qwen, Kimi.
  • Code to prompt: the job you would give Repomix, gitingest, or code2prompt, with a live preview, a token estimate, and secrets masked in place.
  • AI agents: any MCP client reads your project through a server that only reads and always masks, and you steer its focus with checkboxes.
  • CI pipelines: fail a build when secrets would leak into packed context, without ever printing the values.
  • Restricted environments: get the same clean context where AI agents, remote indexing, or IDE plugins are not allowed.
  • Large codebases: pull out one module instead of the whole repository, and fit it into the model's context window.
  • Code reviews and teaching: share the structure and only the files that matter.

Get started

  1. Open or drop a project folder.
  2. Choose folders, files, filters, ignore rules, and output mode.
  3. Preview, then copy, export, run the same workflow from the terminal, or connect your coding agent through MCP → Live context.

Works with any language, repository, or project structure.

Feature overview ✨

Choose and control

  • Smart Ignore: hides build output, dependency folders, and caches, only when it finds evidence.
  • Hide Secrets: masks detected credentials in place, in every export and every MCP response.
  • Hide private data: optional PII redaction for emails, IPs, MACs, phones, and user paths.
  • Code compression: keeps signatures, empties function bodies; pure code gets about 3× smaller.
  • Strip comments and blank lines: syntax-aware cleanup in 20 language packs; source files untouched.
  • File tree with checkboxes, search, and name filters.
  • Git modes: .gitignore, tracked files, staged files, all current changes, or a ref-to-ref diff.

AI agents

  • Read-only MCP server: six tools by default and all nine with --tool-set full, for any MCP client, on the same engine as the desktop app.
  • Live Context: tick files in the desktop app and the agent works with that selection on its very next call.
  • Code search: text, regular expressions, or declarations by name, across the whole selection.
  • Dependency graph: what a file uses and what uses it, from real imports and type references.
  • Git diff review: staged files, current changes, or a ref-to-ref diff, with secrets masked in both versions.
  • Agent journal: a receipt for every session with calls, delivered paths, tokens, and masked values, never file bodies.
  • Hard boundaries: root jail, untrusted-data markers, and an agent that can only narrow what it sees.

Preview and export

  • Live preview (tree, content, or both) with live counters for lines, characters, and a token estimate.
  • Scrollbar markers show search matches and masking findings at a glance.
  • Export as ASCII, Markdown, JSON, or XML, to a file or the clipboard, tree only, content only, or both.
  • Export a clean copy of your project to a folder or ZIP archive.

Workflow and interface

  • Desktop app, Terminal Workspace, CLI, and MCP server: the same engine, four ways to work.
  • Git tools built in: clone by URL, switch branches, update cached copies.
  • Local profiles remember your settings per project.
  • Light, dark, and system themes; platform-native keyboard shortcuts; localization in 20 languages.
  • Stays smooth even on very large folders.

Project copy export

Use File → Export project → To folder… or To ZIP archive… to create a separate copy of your current selection. The same workflow is available through devprojex export project.

Project copies respect your chosen root folders, file types, ignore rules, and checked items; if nothing is checked, the whole current tree is exported. Directory structure, binary files, and included empty folders are preserved. When Hide Secrets or Hide private data is enabled, detected values in text files are replaced, binary files remain unchanged, and the result is intentionally not a byte-for-byte copy, so it may not build or run. The source project is never modified, and the result can't be written inside it.

MCP server 🤖

DevProjex ships a built-in secure Model Context Protocol server, on the official C# MCP SDK, that turns any local folder or Git repository into token-efficient codebase context for AI agents: Claude Code, Codex, Claude Desktop, Cursor, Google Antigravity, VS Code, or any MCP client. It runs on the same engine as the desktop app, the Terminal Workspace, and the CLI. It works as codebase RAG with zero infrastructure: the agent retrieves exactly what a task needs through code search, the dependency graph, and Git diffs, all computed locally from your source files, with no embeddings to build and no vector database to host. One line connects Claude Code, with or without the desktop app installed:

claude mcp add devprojex -- npx -y devprojex mcp --root /absolute/path/to/project

Under the hood every client starts the same command, devprojex mcp --root /path/to/project, and talks to it over stdio.

Connect a client

  • Desktop app: MCP → Live context or MCP → Standard, then Open in Claude Code, Codex, Cursor, or VS Code. The submenu you pick decides whether the registration follows the checked tree.
  • Terminal Workspace: mcp connect <client> [live|standard] registers the server without opening another terminal; live mode is the default here.
  • CLI: devprojex mcp connect . --client <client> (standard mode unless --mode live is given). Add --open

Truncated for display — read the full file on GitHub.

Related Skills

View on GitHub
GitHub Stars44
CategoryAI
Updated12h ago
Forks8

Languages

C#

Trust signals

97/100

From repository metadata: license, adoption, age and documentation. Not a code audit — see the Safety scan above for what the skill file itself contains.

1 info