Skip to main content
Glama

MCPort

MCPort is a local-first MCP tool for working with projects through AI. It lets an AI client read code, search files, understand symbols, inspect Git, edit files, and—when explicitly authorized—run controlled development commands inside selected Workspaces.

Project files, runtime state, and credentials stay on your machine by default. You can use MCP locally or enable OAuth/Bearer Token access for selected Workspaces over the public internet.

中文说明:README.zh-CN.md

Features

  • Read and search project files, source code, images, and directories

  • Code indexing plus definition, reference, Hover, and document-symbol support

  • Git status, diffs, history, and blame

  • Contained file creation, editing, import, copy, move, and deletion

  • Checkpoints, tasks, validation, and recovery for file changes

  • Controlled command execution with allowlists, timeouts, output limits, and local confirmation

  • Local or public MCP access for one or more Workspaces

  • Per-language LSP installation and updates, with custom server support

  • Optional Computer Use with screen, mouse, and keyboard control gated by per-action approval

Related MCP server: Project Navigator MCP Server

Connect an AI client

In MCPort, add a project under Projects, enable MCP for the Workspace, then copy its connection information. Add the copied MCP address to your AI client’s MCP settings. For public access, use the Workspace public address and its configured OAuth or Bearer Token authentication.

Common clients that support MCP include Claude Desktop, Cursor, Windsurf, Cline, Continue, and other MCP-compatible clients. Their settings locations and configuration formats vary; follow the client’s current MCP documentation and paste the address or configuration supplied by MCPort.

After installing or updating an LSP, ask the AI client to call server_info again so it can refresh the available language capabilities.

macOS installation

The current macOS release is unsigned. After dragging MCPort.app to /Applications, run:

xattr -dr com.apple.quarantine "/Applications/MCPort.app"

Only do this for the official download after verifying SHA256SUMS.txt.

Connections

Local access accepts loopback requests by default. Public access can use Cloudflare Tunnel, TryCloudflare, FRP Client, or an external tunnel, with OAuth (recommended) or Bearer Token authentication.

Public Workspaces use routes like:

https://mcp.example.com/w/<workspace>/mcp

Files, code, and LSP

File operations remain inside the selected Workspace after realpath and symlink checks. The built-in code index supports TypeScript/JavaScript, Python, Go, Rust, Java, C/C++, and PHP.

Language servers are managed independently and can be installed or updated on demand. Supported languages include TypeScript/JavaScript, HTML, CSS/SCSS/LESS, Python, JSON, YAML, Markdown, Go, Rust, Java, C, C++, and PHP. Custom LSP definitions can be added for other languages.

Permission tiers

  • readonly: project browsing, search, code understanding, read-only Git, and history

  • standard: file changes, imports, checkpoints, tasks, and quick validation

  • full: controlled commands, command sessions, recovery, and complete validation

Commands use an executable name plus an argument array, never an arbitrary shell string. They remain subject to the global command switch, exact allowlist, timeouts, output limits, and high-risk confirmation.

Computer Use is disabled by default and initially available only on local MCP connections in the full tier. A separate switch can expose it to authenticated public Workspaces using the full tier. Screenshots and desktop input still require explicit MCPort Desktop confirmation for every action.

Security

Workspace boundaries are checked after resolving real paths and symlinks. File changes support SHA256 baselines, transactional rollback, and checkpoints. Public Workspaces always require authentication. Tokens, tunnel credentials, and other secrets use system secure storage and are redacted from logs and traces.

The Runtime runs as the logged-in OS user. The command allowlist is not an operating-system sandbox.

Documentation

Contributing

MCPort is maintained by Sky and released under the MIT License. Issues and pull requests are welcome. See CONTRIBUTING.md, SECURITY.md, CODE_OF_CONDUCT.md, privacy notes, and third-party notices.

Third-party LSPs are installed from their own official package managers or release channels and are not bundled with MCPort. Follow each project’s license and terms.

A
license - permissive license
Not graded
quality - not tested
A
maintenance

Maintenance

Maintainers
Response time
0dRelease cycle
2Releases (12mo)
Commit activity

Resources

Unclaimed servers have limited discoverability.

Looking for Admin?

If you are the server author, to access and configure the admin panel.

Related MCP Servers

View all related MCP servers

Related MCP Connectors

  • Securely search and manage workspace context files for AI agents and teams.

  • Git-backed platform for skills, tools, and context for AI agents

  • Cross-agent artifact workspace with provenance across Claude Code, Codex, Cursor, LangGraph.

View all MCP Connectors

Latest Blog Posts

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/sky32/mcport'

If you have feedback or need assistance with the MCP directory API, please join our Discord server