Verified Repo Memory
Give agents repo-scoped memory that cites sources, re-verifies facts on demand, and expires stale entries so long sessions do not rot context.
Overview
Verified Repo Memory is a MCP server for the Build phase that keeps repo-scoped agent memory fresh with citations, just-in-time verification, and TTL expiry.
What is this MCP server?
- Stale-proof repository memory scoped per repo
- Citations on stored facts for agent-auditable answers
- Just-in-time verification before trusting recalled items
- Configurable TTL (VRM_TTL_DAYS 1–365) and data dir overrides
- Optional secret scan (VRM_SECRET_SCAN) and max file parse size (VRM_MAX_FILE_BYTES)
- npm package @cognitivemyriad/vrm-local version 0.1.2
- VRM_TTL_DAYS configurable from 1 to 365 days
- Four documented environment variables in server metadata
Community signal: 1 GitHub stars.
What problem does it solve?
Coding agents accumulate wrong repo ‘facts’ that linger across sessions because nothing re-checks files or expires old notes.
Who is it for?
Solo builders on long-lived repos who want local, verifiable MCP memory instead of unbounded chat history.
Skip if: Teams that only need ephemeral chat with no persisted memory, or cloud-only knowledge bases with no repo tie-in.
What do I get? / Deliverables
Agents recall project memory with citations and freshness controls so answers stay aligned with the current tree.
- Persisted repo memory entries with citation metadata
- JIT verification passes before agents rely on recalled facts
- TTL-driven expiry to limit stale architectural assumptions
Recommended MCP Servers
Journey fit
Persistent, verifiable repo memory is agent-tooling infrastructure you add while building how your coding agent remembers the codebase. Citations, JIT verification, and TTL belong next to other local MCP memory layers—not in launch or grow marketing work.
How it compares
Repo-scoped verified memory MCP, not a marathon calculator or Linux desktop automation suite.
Common Questions / FAQ
Who is Verified Repo Memory for?
Developers using MCP coding agents who need durable, citation-backed repository memory that can expire and be re-verified.
When should I use Verified Repo Memory?
Use it while setting up agent-tooling on a codebase you will touch for weeks, especially when stale architecture notes have burned you before.
How do I add Verified Repo Memory to my agent?
Install @cognitivemyriad/vrm-local, configure stdio in your MCP client, and set optional env vars such as VRM_TTL_DAYS, VRM_DATA_DIR, VRM_MAX_FILE_BYTES, and VRM_SECRET_SCAN.