by mcpdelta
Delta MCP is a free app that sits between your AI apps and their MCP servers: one program per task instead of one tool call per step, up to 24.1× fewer tokens in our benchmarks. Every change is recorded and undoable. Coming soon for macOS, Windows and Linux.
# Add to your Claude Code skills
git clone https://github.com/mcpdelta/mcpdeltaSee how mcpdelta compares with popular alternatives.
mcpdelta is an open-source ai agents skill for AI coding assistants such as Claude Code, Codex CLI, and ChatGPT, built by mcpdelta. Delta MCP is a free app that sits between your AI apps and their MCP servers: one program per task instead of one tool call per step, up to 24.1× fewer tokens in our benchmarks. Every change is recorded and undoable. Coming soon for macOS, Windows and Linux. It has 79 GitHub stars.
mcpdelta's catalog security scan is still queued. You can run an instant dependency and prompt-injection check now with the "Scan for vulnerabilities" button above.
Clone the repository with "git clone https://github.com/mcpdelta/mcpdelta" and add it to your Claude Code skills directory (see the Installation section above).
Yes. SkillsLLM lists many other AI Agents skills you can browse and compare side by side. Open the AI Agents category from the badge at the top of this page, or use the Related Skills and comparison links further down to weigh mcpdelta against similar tools.
No comments yet. Be the first to share your thoughts!
⚠️ Third-Party Software Notice
This skill is third-party open-source software developed and hosted independently on GitHub. SkillsLLM is an informational directory and does not control or maintain the underlying repository.
Any security checks, ratings, or warnings displayed by SkillsLLM are automated and limited in scope. They do not constitute a security certification or guarantee that the software is safe, error-free, or free from malicious code, vulnerabilities, compromised dependencies, or prompt-injection risks.
Review the source code, permissions, dependencies, and configuration before installing or running any third-party skill. Use is at your own risk. To the maximum extent permitted by applicable law, SkillsLLM is not liable for losses arising from third-party software.
The deep catalog scan for this skill is still queued. Run an instant dependency check now instead.
Delta MCP is a free app that sits between your AI apps and their MCP servers. Your AI does each task as one program instead of one tool call per step. Delta MCP checks it before the first call, makes the fewest calls, all or nothing, reads the result back, records it and can undo it. It works with the MCP servers you already use.
Coming soon. Delta MCP is almost here, for macOS, Windows and Linux: free, no account. Follow the launch and you’ll know the moment it’s out, with no email address: RSS, or Watch → Custom → Releases on this repository.
Your AI becomes a developer, not an executor. Instead of reading, deciding, calling one tool, waiting and reading again, it writes the whole task once.
Same result, up to 24.1× fewer tokens. With Delta MCP, multi-step tasks use far fewer tokens, cost less and finish as fast or faster, with the same end result, checked in the service itself. 8 scenarios, each run with and without Delta MCP, in Claude Code with Sonnet 5.5.
| 3.3–24.1× | 1.4–8.1× | up to 4.4× | 152 / 152 |
|---|---|---|---|
| fewer billed tokens on multi-step tasks | lower cost on multi-step tasks, even in the strictest cache case | faster on multi-step tasks, and no scenario is slower | tools still declared and callable through Delta MCP |
| Task | Server | Billed tokens | Model calls |
|---|---|---|---|
| Fix a release, as text | Cerberus, 152 tools | 24.1× fewer | 26 → 4.5 |
| Add tests, as text | Cerberus, 152 tools | 15.8× fewer | 16.6 → 4 |
| Fix a release, as code | Cerberus, 152 tools | 13.4× fewer | 26 → 6.5 |
| Clean up 20 late tickets | A ticket manager | 4.8× fewer | 5 → 2 |
| Log in, add 3 contacts, list them | Microsoft Playwright MCP | 3.9× fewer | 12.6 → 4 |
| Make 6 changes to a task list | taskqueue-mcp | 3.3× fewer | 5.6 → 2 |
| A documentation question | Context7 | 1.2× fewer | 4 → 3 |
| A one-step question | Playwright, books.toscrape.com | 1.25× more | 4 → 3 |
Averages of 2 to 3 runs per variant. A single one-step question costs more, because your AI reads Delta MCP’s short description on every call: the gain is on tasks with several steps. A benchmark on the biggest MCP servers will be published when Delta MCP is released.
All benchmarks, with the method →
Claude Desktop, Claude Code, Cursor, VS Code, Windsurf, Gemini CLI, Cline, Roo Code, LM Studio, Amazon Q, Codex and Zed. Delta MCP switches the first ten over for you; Codex and Zed are set up by hand for now. Your MCP servers stay exactly as they are. See how each app is switched over.
The free app for macOS, Windows and Linux is coming soon. There is nothing to install yet, and this repository does not contain the app.
| Option | Where |
|---|---|
| Download page | mcpdelta.com/download |
| Follow with RSS | mcpdelta.com/feed.xml |
| Watch releases here | Watch → Custom → Releases, on this repository |
Delta MCP is a free app that sits between your AI apps and their MCP servers. Your AI does each task as one program instead of one tool call per step. Delta MCP checks it before the first call, makes the fewest calls, all or nothing, reads the result back, records it and can undo it. It works with the MCP servers you already use.
Yes. Every feature is in the free app, with no account and no limit on servers. See pricing.
No. Delta MCP is a free app, and its source code is not public. How it works is written out in the docs, how it is measured is on the benchmarks page, and everything it does on your computer is recorded in Activity. This repository only presents the app.
Delta MCP works with 12 AI apps: Claude Desktop, Claude Code, Cursor, VS Code, Windsurf, Gemini CLI, Cline, Roo Code, LM Studio, Amazon Q, Codex and Zed. It switches the first ten over for you; Codex and Zed are set up by hand for now.
Yes, on tasks with several steps: 3.3 to 24.1 times fewer tokens in our measurements. A single one-step question costs more (about 1.25 times the tokens), because your AI reads Delta MCP’s short description on every call.
No. Delta MCP runs on your computer and talks to your servers directly; your tools, tasks and history never leave it. Besides a check for new versions, which sends nothing about you, the only thing that can be sent is one anonymous number a day, your tokens saved: you choose at first launch, and it’s off in Settings in one click. The complete list is on the security page.
No. Delta MCP uses them exactly as they are. Every original tool stays callable by name: 152 of 152 on the largest server we tried.
Yes, on tasks with several steps. Tool Search keeps the list of tools short, but every step is still a separate request to the model. Delta MCP removes those requests: your AI writes the task as one program. Our measurements were taken with Tool Search on, so the savings come on top of it.
It is closer to a local proxy than to an enterprise gateway. Like a proxy, it puts your MCP servers behind one entry in your AI apps. Unlike a gateway, it is a free app on your own computer, with no service to run. Its main job is different: your AI writes each task as one program, so it makes far fewer calls.
Settings, then Put everything back. Each connector returns to your AI apps exactly as it was; Delta MCP keeps a copy of every file it changed.
Soon. Follow the launch with RSS or Watch → Custom → Releases on this repository, and you’ll know the moment it’s out.