claude-desktop-extensions
Skill5 starsChanged yesterdayMITView on GitHub
Build, validate, pack, sign, and distribute Claude Desktop Extensions (.mcpb, formerly .dxt): a ZIP of a local MCP server plus manifest.json that users install with one click in Claude Desktop. Use when the user wants to package an MCP server for Claude Desktop, write an MCPB manifest.json, use the mcpb CLI, handle user_config or signing, or submit to the extension directory. Not for mods or plugins.
- Installs packages
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 0 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 43
Config, not constants: put keys, paths, and options in user_config and reference as ${user_config.KEY} in server.mcp_config.
≈ “Common” is an estimate: at least 1 in 5 of the 308,232 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Claude Desktop Extensions (MCPB)
An MCPB (.mcpb, formerly .dxt; .dxt still works, .mcpb preferred) is a ZIP containing a local MCP server and a manifest.json. manifest.json is the only required file. Claude Desktop ships Node.js, so users need no runtime setup for node servers. Spec and CLI: https://github.com/modelcontextprotocol/mcpb (MANIFEST.md, CLI.md). Manifest spec is 0.3; the uv type is shown under "manifest_version": "0.4". Check the repo's MANIFEST.md for the current version before writing.
Workflow
- Confirm the fit: a local server the user runs on their machine. For a hosted server, don't package.
- Scaffold:
mcpb initin the server folder (interactive), or hand-writemanifest.jsonfrom references/manifest.md. - Pick the server type:
node(bundlenode_modules),python(bundle deps inserver/liborserver/venv),binary(compiled;.exeappended on Windows; useplatform_overrides), oruv(needspyproject.toml; must not includeserver/liborserver/venv). - Declare tools in
tools[](andprompts[]if any) so Claude Desktop can show them before launch. Usetools_generated: truewhen the list is dynamic. - Config, not constants: put keys, paths, and options in
user_configand reference as${user_config.KEY}inserver.mcp_config. Mark secretssensitive: true(stored in the OS keychain). Claude Desktop will not enable the extension until required values are set. - Validate:
mcpb validate manifest.json. - Pack:
mcpb pack <dir> [out.mcpb](validates, excludes dev files; add a.mcpbignorefor more). Check the result withmcpb info out.mcpb. - Test on real installs: drag the
.mcpbinto Claude Desktop Settings -> Extensions (or double-click it). Test on Windows and macOS if both are incompatibility.platforms. - Sign for distribution:
mcpb sign(PKCS#7).--self-signedis for development only; production needs a code-signing certificate. Verify withmcpb verify. - Distribute: share the file, host it, submit to Anthropic's extension directory (form; test both OSes first), or roll out to a fleet with GPO/MDM, preinstall, blocklists and private directories.
Manifest essentials
Required: manifest_version, name, version (semver), description, author { name, email?, url? }, server.
Recommended: display_name, long_description, icon/icons, screenshots, repository, homepage, documentation, support, keywords, license, compatibility, and privacy_policies[] (required when the server connects to external services handling user data).
server: type, entry_point, mcp_config { command, args, env, platform_overrides }.
Variables available in mcp_config: ${__dirname}, ${HOME}, ${DESKTOP}, ${DOCUMENTS}, ${DOWNLOADS}, ${pathSeparator} (or ${/}), ${user_config.KEY}.
Complete field tables and a working example: references/manifest.md. CLI, signing and distribution: references/cli-and-distribution.md.
Security rules
- Extensions run local code with the user's file and network access. Request the minimum; document what it reads and sends.
- Never hardcode credentials, bake secrets into the bundle, or log them. Use
user_config+sensitive: true. - Validate and sanitize every tool argument; no shelling out with unescaped input; constrain file access to the declared
directory/fileconfig. - Provide a privacy policy URL if user data leaves the machine.
- Sign production builds; do not ship self-signed to others.
- Pin dependencies; bundle only what runs (
.mcpbignoreout tests, docs, caches).
Verification checklist
mcpb validatepasses andmcpb packproduced a.mcpbmcpb infoshows the expected name, version, and sizeserver.typematches what is bundled (anduvhas noserver/lib/server/venv)- All secrets come from
user_configwithsensitive: true; none in the zip compatibility(claude_desktoprange,platforms,runtimes) is accurate- Installed in Claude Desktop (Settings -> Extensions) and tools actually appear and run
- Windows and macOS tested if both are declared
privacy_policiespresent when external services are used- Production build signed with a real certificate
Anti-patterns
- Wrong
manifest_versionfor the features used. - Using absolute or machine-specific paths instead of
${__dirname}/ user variables. - Relying on a global
node/pythoninstall instead of the declared type and bundled deps. - Bundling the entire repo (tests,
.git,node_modulesdev deps). - Treating
--self-signedas production. - Packaging a hosted/remote server as a
.mcpb. - Guessing manifest fields instead of reading
MANIFEST.mdfor your version.
Pointers
| Need | Read |
| manifest.json fields and example | references/manifest.md |
| mcpb CLI, signing, directory, enterprise | references/cli-and-distribution.md |
Bundling the .mcpb inside a plugin (mcpServers) | claude-desktop-plugins |
In this repository30 other files it gives its agents.
- docs/agents.mdAGENTS.md
- website/AGENTS.mdAGENTS.md
- .github/copilot-instructions.mdCopilot instructions
- .cursor/rules/claude-desktop-skills.mdcCursor rule
- .cursor/rules/claude-mod.mdcCursor rule
- .cursor/rules/codex-desktop-skills.mdcCursor rule
- .cursor/rules/desktop-and-publishing-skills.mdcCursor rule
- .cursor/rules/release.mdcCursor rule
- claude-code-mods-guideSkill
- claude-desktop-modsSkill
- claude-desktop-pluginsSkill
- codex-desktop-extensionsSkill
- codex-desktop-pluginsSkill
- codex-desktop-skillsSkill
- copilot-desktop-extensionsSkill
- cursor-extensions-and-pluginsSkill
- opencode-pluginsSkill
- publishing-mental-to-storesSkill
- claude-code-mods-guideSkill
- claude-desktop-extensionsSkill
- claude-desktop-modsSkill
- claude-desktop-pluginsSkill
- codex-desktop-extensionsSkill
- codex-desktop-pluginsSkill
- codex-desktop-skillsSkill
- copilot-desktop-extensionsSkill
- cursor-extensions-and-pluginsSkill
- opencode-pluginsSkill
- publishing-mental-to-storesSkill
- mental-setupSkill
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
Reviews can't be read right now.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

