Portsmith
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
Portsmith
Move JS/TS features between repos—with behavior checks and a reviewable patch.
Quick start · First transfer · Coding agents · Examples · Discussions
Found a useful feature in another codebase? Portsmith traces its JS/TS files, stages them in a copy of your app, and checks the integration against behavior you declare. Use your own edits, a coding agent, or an OpenAI-compatible model. Your original repositories stay unchanged.

Recorded from the source demo using an included authored proposal, not autonomous AI. The published alpha has the same example behavior with simpler browser styling. Demo details and reproduction · Still image
Try it now
Node 22.19+ (22.x) or 24.11+, plus Git. No API key or signup needed:
npx --yes --package=portsmith-workbench@alpha portsmith-demoThe included example moves table filtering into a catalog app:
Source contract 5/5 passed
Destination before 1/5 passed
Destination after 5/5 passed
Existing tests pass before and after
Output a patch + verification reportIt creates a fresh .portsmith/demo-* folder in your current directory and prints the patch and report paths. The proposal is authored so you can try the workflow without a model.
For ongoing use:
npm install -g portsmith-workbench@alpha
portsmith doctor --human
portsmith-demo --install-browserThe last command downloads matching Chromium and runs browser interaction checks. Later use portsmith-demo --browser. Minimal Linux hosts may also need npx playwright@1.61.1 install-deps chromium. Installation and troubleshooting →
Related MCP server: Hoplon
What can I use it for?
Reuse a focused feature: move a filter, parser or transformation into another JS/TS app.
Review an agent's integration: test the proposed behavior, retain existing tests, then inspect the patch.
Check a browser workflow: declare interactions such as search and reset, and retain screenshots and traces.
Start with a small feature whose inputs, outputs and destination tests you can name. Walk through a complete transfer →
How it works
Discover → Inspect → Declare behavior → Prepare a copy
↓
Review patch ← Export ← Verify ← Adapt the integrationExport requires a passing source, a demonstrated destination gap, a passing candidate, completed regressions, unchanged originals and a fresh patch that reproduces the verified candidate. A failure, timeout or skipped test blocks verification. Full verification rules →
Bring your coding agent
Portsmith includes a CLI, MCP server and typed JavaScript API. Connect an MCP host such as Claude Code or Cursor, then ask:
Use Portsmith to find the search feature in
/work/sourceand plan a transfer into/work/app. Show me the behavior contract and dependencies before executing code. Preserve existing tests and export a patch I can review.
Copyable MCP configurations · Agent workflow skill · API reference
For Portsmith's own model-driven adaptation:
portsmith init --out ./portsmith-setup --model YOUR_MODEL_ID
# Copy portsmith-setup/.env.example to .env in that directory; add your key locally.Then follow the model setup guide. Your key stays in the environment; model use sends selected code to your configured provider. The no-key demo above does not exercise autonomous adaptation.
Scope of this alpha
Works within declared scope | Limits to know |
JS/TS file tracing and feature discovery | Lexical leads, not runtime proof or function slicing |
npm/pnpm setup in fresh copies, including workspaces | Scripts disabled; unknown versions need reviewed declarations |
Static aliases and explicit Vite/plugin resolution captures | Computed imports and arbitrary bundler assets are unsupported |
Function contracts, Node/Vitest regressions, declared builds | Only completed declared checks count |
Chromium interactions, screenshots and traces | No general cross-framework conversion or production guarantee |
Public alpha: 0.6.0-alpha.2. Review every patch. Verification executes repository code on your machine; it is not a security sandbox. Use trusted code and review source licenses. A pass proves the declared cases, not general equivalence. Dependency details · Security
Examples and evidence
Historical controlled transfers include IT-Tools, TodoMVC, drawDB and Mermaid Live Editor. They have authored boundaries and recorded limitations; they are not a general success rate or proof of parity with Rea. Browse the cases →
Rea is an optional analysis evidence adapter. Its findings stay separate from Portsmith's behavior checks. Adapter and comparison scope.
Documentation and contributing
I want to… | Start here |
Install or diagnose setup | |
Understand every step | |
Use an agent or a model | |
Transfer a larger app | |
Declare browser checks | |
Find commands or API types | |
Contribute or report a problem |
git clone https://github.com/punsnprotons/portsmith.git
cd portsmith
npm ci --ignore-scripts
npm run browser:install
npm run check
npm run demoThis server cannot be deployed
Maintenance
Related MCP Connectors
Team memory for AI coding agents and code reviews, reviewed by humans and served over MCP.
Hosted code graph over MCP: exact callers, dependencies, and cross-repo blast radius for AI agents.
Deploy & release tracking with native MCP — ask Claude what's in production and cut the release.
AI-native git hosting — repos, PRs, issues, CI gates, and AI code review over MCP (60 tools).
Related MCP Servers
- FlicenseNot gradedqualityCmaintenanceAnalyzes repositories, explains architecture, calculates change impact, and enforces guardrails for AI Agents like Claude Code, Cursor, and Codex via MCP tools.-
- AlicenseNot gradedqualityCmaintenanceEnables AI coding agents and hosts to enforce deterministic repository boundaries via MCP, providing structured reads, supervised edits, snapshots, audits, and recovery with machine-readable evidence.MIT
- AlicenseNot gradedqualityCmaintenanceEnables AI coding agents to search indexed local Git repositories for compatible existing projects and modules, evaluate their technical and structural fit, and copy reusable source bytes into a new target instead of regenerating them. It exposes six scoped MCP tools that return plans and safe file operations, so the agent only implements the remaining delta.1MIT
- AlicenseNot gradedqualityAmaintenanceEnables MCP-capable coding agents to run a human-gated, evidence-backed reverse-engineering pipeline that turns an unfamiliar repository into layered architecture, behavioral contracts, defect findings, validation gates, and a language-agnostic reimplementation spec.444 npm7MIT