Mneme Decision MCP — Claude Code MCP server: what it does
A local stdio MCP server for project architecture decisions and ADRs. It gives coding agents context on engineering standards while keeping approval authority strictly human.
claude mcp add mneme -- uvx mneme-hqWhat Mneme Decision MCP does
Mneme Decision MCP connects coding agents to the decision index stored in .mneme/project_memory.json. Agents query established design choices before editing files and verify compliance against recorded standards. Python 3.11+ is required.
The server surface contains exactly six tools: decision.propose, decision.propose_batch, decision.get, decision.search, decision.applicable_to, and decision.trace. The agent can submit candidate proposals or look up previous choices. Crucially, the server exposes zero tools for accepting, superseding, or bypassing decisions, keeping final governance entirely in human hands via the CLI.
Startup fails closed if .mneme/project_memory.json lacks an authoritative decision_index block. In that case, you must run mneme decision-index migrate before the server will answer queries.
What it helps with
- query architectural decisions that apply to specific modules or source files
- submit draft architectural proposals directly from coding agents
- trace the origin and context behind past engineering choices
- prevent unintended architecture drift during automated code edits
Example prompts for the agent
- Check Mneme for architectural decisions that apply to our database layer.
- Search project memory for decisions regarding authentication standards.
- Propose a draft architectural decision for caching API responses.
How to install Mneme Decision MCP
Platform CLI
claude mcp add mneme -- uvx mneme-hquvx runs the command named after the package; if the README names another one, use uvx --from mneme-hq <command>
Platform CLI
codex mcp add mneme-decision-mcp -- uvx mneme-hquvx runs the command named after the package; if the README names another one, use uvx --from mneme-hq <command>
Config file
{
"mcpServers": {
"mneme-decision-mcp": {
"command": "uvx",
"args": [
"mneme-hq"
]
}
}
}uvx runs the command named after the package; if the README names another one, use uvx --from mneme-hq <command>
Config file
{
"servers": {
"mneme-decision-mcp": {
"type": "stdio",
"command": "uvx",
"args": [
"mneme-hq"
]
}
}
}uvx runs the command named after the package; if the README names another one, use uvx --from mneme-hq <command>
Platform CLI
gemini mcp add mneme-decision-mcp uvx mneme-hquvx runs the command named after the package; if the README names another one, use uvx --from mneme-hq <command>
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 Mneme Decision MCP
How do I install Mneme Decision MCP in Claude Code?
Run claude mcp add mneme-decision-mcp -- uvx mneme-hq in your terminal.
What does Mneme Decision MCP do?
It provides AI coding agents with read access to project architectural decisions and ADRs. Agents can also submit proposals, but all enforcement state remains inspectable and controlled.
Can an AI agent approve its own architectural proposals?
No. The MCP server deliberately lacks accept, activate, or bypass endpoints, so humans must review and accept proposals using mneme decision accept in the CLI.
