stack MCP server
Commit Oct 26, 2026Provides a lifecycle toolkit and pre-flight checklist for building, testing, and publishing MCP servers.
Also known as anand2426/mcp-stack
- Stars
- 0
- Tools
- —
- Downloads, 30 days
- —
- Latest version
- —
- Last commit
- Oct 26, 2026
- License
- MIT
What is stack MCP?
stack provides a lifecycle toolkit and pre-flight checklist for building, testing, and publishing MCP servers.
Summary generated from the project's README
About stack MCP server
Developers building Model Context Protocol plugins can use this toolkit to validate their work before it reaches users. It functions as a Claude Code plugin that provides a structured lifecycle for creating, testing, and releasing servers. Instead of starting from scratch, you can use the scaffold skill to generate TypeScript or Python skeletons from templates. The toolkit focuses on catching errors like vague tool descriptions, overbroad permissions, or leaked secrets that often occur during rapid development.
Users interact with the system through the Claude Code interface by typing commands or describing their intent. The toolkit includes specialized skills like mcp-stack-lint-schema for clarity, mcp-stack-security-review for safety, and mcp-stack-eval-harness to verify that tool responses match their definitions through real connections. It also manages the final stages of the process with mcp-stack-publish for versioning and changelog creation, alongside mcp-stack-canary to monitor for silent breaks caused by dependency changes after a release.
What you can do with it
- 1Generate a new server templateCreate a working TypeScript or Python server skeleton instead of starting with a blank file.
- 2Audit tool descriptions for clarityCheck if tool parameters are too ambiguous for an AI agent to use without human intervention.
- 3Perform a security auditIdentify overbroad permissions, leaked secrets, or unsafe tool defaults like unguarded shell access.
- 4Verify end-to-end tool correctnessTest that the server tools actually return the promised responses through a real MCP connection.
Written from the project's README. Every use case is checked against a quote from it.
README
View on GitHubmcp-stack
A pre-flight checklist and lifecycle toolkit for building and publishing MCP servers — the plugins that let an AI agent talk to a new tool (Gmail, a database, your own API).
The problem: MCP servers are easy to write and fast to publish, so most of them ship same-day with no one checking that the tool descriptions are clear, the permissions aren't overbroad, no secret got committed, and the server actually works end-to-end with a real agent. mcp-stack is the checklist that catches that before publishing — not after.
What it is
A Claude Code plugin — install it and you get a set of skills covering the whole lifecycle of an MCP server:
| Stage | Skill | What it checks |
|---|---|---|
| Start | mcp-stack-scaffold |
Generates a working server skeleton (TypeScript or Python) from a template, not a blank file. |
| Clarity | mcp-stack-lint-schema |
Flags tool descriptions/parameters too vague or ambiguous for an agent to use correctly, no human in the loop to ask. |
| Safety | mcp-stack-security-review |
Flags overbroad permissions, leaked secrets, and unsafe tool defaults (unguarded shell/write access). |
| Correctness | mcp-stack-eval-harness |
Actually calls the server's tools through a real MCP connection and checks the responses match what's promised — not just that the code compiles. |
| Ship | mcp-stack-publish |
Version bump, changelog, registry submission — only on your explicit go-ahead, never automatic. |
| Watch | mcp-stack-canary |
Ongoing drift checks after publishing, so a later dependency/API change doesn't break it silently. |
Each stage can also hand off to a reviewer subagent (agents/) for a deeper pass: schema-quality-reviewer, security-reviewer, dx-reviewer.
Install
# as a local plugin while developing it
claude --plugin-dir ./mcp-stack
# once published to a marketplace
/plugin install mcp-stack@<marketplace-name>
Use
/mcp-stack
or just describe what you want ("lint my MCP server's tool schemas", "security-review this before I publish") and the router skill hands off to the right stage.
Repo layout
mcp-stack/
.claude-plugin/plugin.json # plugin manifest
skills/ # one SKILL.md per lifecycle stage
agents/ # reviewer subagents invoked by the skills
scripts/ # standalone checkers the skills shell out to
templates/server-skeleton/ # scaffold output for mcp-stack-scaffold
Status
v0.1.0 — early. lint_schema.py and secret_scan.py are working standalone scripts; the rest of the lifecycle (eval harness, publish, canary) is currently agent-driven via the SKILL.md instructions rather than scripted. Contributions and real-world test reports welcome.
License
MIT — see LICENSE.
stack MCP questions
What does mcp-stack-scaffold do?
It generates a working server skeleton in either TypeScript or Python using a template.
How does the tool ensure the server works correctly?
The evaluation harness calls the server's tools through a real MCP connection to verify responses match the promises.
Does the publishing process happen automatically?
No, publishing only occurs with your explicit permission and never happens automatically.
What is the current development status of the project?
The project is in version 0.1.0, where schema linting and secret scanning are working standalone scripts, while other stages are agent-driven.