mcpcall keeps the AI tooling you have spent years refining out of the client repos you push to. Author it once on the platform, pull it into any local checkout over MCP, and leave nothing behind when the engagement ends.
{
"mcpServers": {
"mcpcall": {
"command": "npx",
"args": ["-y", "@asad120414/mcpcall"],
"env": {
"MCPCALL_API_KEY": "${MCPCALL_API_KEY}"
}
}
}
}$ pull_workspace acme-corp
A shared .gitignore is not a security boundary.
You consult. You do not own the remote. Years of subagents, skills, and per-client conventions live on your machine — and the standard workaround for keeping them there is to copy markdown into the project and add the names to .gitignore.
That file is tracked. One git add -A away from leaking. The file names alone tell the client what you brought. And when the engagement ends there is no clean off-boarding.
mcpcall replaces the workaround with infrastructure. The files live on the platform. They land on your disk only when you ask for them, hidden from git through .git/info/exclude — user-local, never tracked, never pushed. Cleanup removes every byte.
Write CLAUDE.md, subagents, skills, and slash commands in a workspace per client or project.
Drop a small .mcp.json into the client repo. The mcpcall server boots via npx — no global install.
In Claude Code, run pull_workspace <slug>. Files land in .claude/, hidden via .git/info/exclude.
One call. Sha-matched files removed, sentinel block stripped, repo is bit-identical to before.
Everything needed to round-trip context files between the platform and any local checkout — no clobbering, conflict-safe, fully reversible.
list_workspaceslist_filesstatuspull_workspacepull_workspace acme-corppull_filecreate_filepush_fileimport_folderdelete_filecleanupFull reference in the tool docs.
Open signup. Email or GitHub, your call. Free during beta — no card, no quota.