m1nd — Claude Code MCP server: a code graph your agent can check
A local code graph with evidence-linked memory for MCP agents: orient in an unfamiliar repo, see what a change touches, keep verified findings for the next session.
claude mcp add --env M1ND_WORKSPACE_ROOT=<M1ND_WORKSPACE_ROOT> --transport stdio m1nd -- npx -y @maxkle1nz/m1ndWhat m1nd does
The server builds a graph of your repository on your machine and exposes it to an MCP host. Your agent can ask where configuration is read, which files connect to a change, and what sits around a symbol. The north, seek and impact tools return candidates along with source excerpts, a trust level and the gaps left open. A link in the graph tells the agent where to look; it never proves an edit is safe.
memorize keeps a claim next to the paths that back it, so a later session can pull it up and check whether it still holds. reverify, abstain and the insufficient_evidence result are ordinary answers here, not failures. A local owner can serve one graph to attached clients; the presence and overlap warnings help agents avoid stepping on each other but lock nothing.
Graph and memory stay on disk, and graph queries need no LLM API key. You need Node.js 18+ and a readable repository; the native runtime binary comes from the signed release instructions. Set M1ND_WORKSPACE_ROOT to the absolute path yourself, because hosts apply leaves that selector out and a host started elsewhere will open a different graph. Installation, updates and network-backed document providers may reach the network, and your MCP host applies its own data policy.
What it helps with
- locate where configuration gets loaded
- inspect a function's neighbours before editing
- store a verified finding with the paths behind it
- recall earlier findings in a new session and recheck them
- tell a confident answer from a missing one
Example prompts for the agent
- Check with m1nd which files a change to
config/loader.tswould touch. - Pull my earlier m1nd findings on this module and reverify whether they still apply.
How to install m1nd
Platform CLI
claude mcp add --env M1ND_WORKSPACE_ROOT=<M1ND_WORKSPACE_ROOT> --transport stdio m1nd -- npx -y @maxkle1nz/m1ndReplace <…> with your own values
Platform CLI
codex mcp add m1nd --env M1ND_WORKSPACE_ROOT=<M1ND_WORKSPACE_ROOT> -- npx -y @maxkle1nz/m1ndReplace <…> with your own values
Config file
{
"mcpServers": {
"m1nd": {
"command": "npx",
"args": [
"-y",
"@maxkle1nz/m1nd"
],
"env": {
"M1ND_WORKSPACE_ROOT": "<M1ND_WORKSPACE_ROOT>"
}
}
}
}Replace <…> with your own values
Config file
{
"servers": {
"m1nd": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"@maxkle1nz/m1nd"
],
"env": {
"M1ND_WORKSPACE_ROOT": "<M1ND_WORKSPACE_ROOT>"
}
}
}
}Replace <…> with your own values
Platform CLI
gemini mcp add -e M1ND_WORKSPACE_ROOT=<M1ND_WORKSPACE_ROOT> m1nd npx -y @maxkle1nz/m1ndReplace <…> with your own values
Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about m1nd
How do I install m1nd in Claude Code?
Run claude mcp add m1nd -- npx -y @maxkle1nz/m1nd. To pin the workspace root explicitly, use the stdio form: claude mcp add --env M1ND_WORKSPACE_ROOT=<M1ND_WORKSPACE_ROOT> --transport stdio m1nd -- npx -y @maxkle1nz/m1nd.
What does m1nd do?
It keeps a local graph of your code alongside memory entries tied to evidence paths. The agent finds connected code, inspects what a change reaches, and carries checked findings forward.
Does m1nd need an LLM API key?
No for graph queries. Installation, updates and document providers that forward documents to a configured endpoint may use the network.
