en-quire

Structured document management for agent systems, with governance.

An MCP server that treats markdown and YAML files as structured, section-addressable documents with built-in RBAC, approval workflows via git, and semantic search. Designed for operational use cases where agents need to read, propose edits to, and maintain documents — SOPs, skill files, memory, runbooks, config files — under governance.

> A Nullproof Studio open-source project.

---

Agent systems increasingly depend on markdown files as operational infrastructure. But existing tooling falls short:

• Filesystem MCP — no document awareness, no governance, no search. Agents can clobber files freely.
• Knowledge-graph MCPs — impose opinionated schemas, designed for conversational memory rather than operational documents.
• Search-only MCPs — read-only. No write or edit capability.
• None of them have RBAC or approval flows. Every caller is fully trusted.

en-quire fills this gap: a server that understands document structure, supports surgical section-level editing, and treats governance as a first-class concern.

• Multi-format support — pluggable parser architecture handles markdown (.md, .mdx) and YAML (.yaml, .yml). Both produce the same section tree; all tools work uniformly across formats.
• Section-addressable editing — read and write at the heading/key level, not the file level. Address sections by heading text (## Checks), breadcrumb path (Procedures > Checks > Daily), positional index ([0, 1]), or YAML dot-path (services.api.environment.PORT).
• Multi-root document management — configure multiple named document roots with independent git repos, permissions, and search indices. Paths are prefixed by root name (docs/sops/runbook.md, config/docker-compose.yaml).
• Git-native governance — edits from unprivileged callers land on branches, not main. Approval i