@doquflow/server

MCP server that lets AI agents read any codebase and write persistent specs. Zero AI inside — the agent does all the thinking.

AI Agent (Claude Code)
       │ calls MCP tools
@doquflow/server          ← reads files, extracts, writes markdown
       │ filesystem only
Any project on disk       ← any language, any structure

Install

npx @doquflow/cli init

The CLI registers this server automatically in your Claude Desktop config. You do not need to install @doquflow/server directly.

Tools

| Tool | Input | Output | |------|-------|--------| | read_module | { path } | Classes, functions, deps, endpoints, tables, config refs, raw content | | list_modules | { path, extensions? } | All modules in a directory tree | | write_spec | { project_path, filename, content } | Writes a markdown spec to .docuflow/specs/ | | read_specs | { project_path, module_name? } | Reads existing specs from .docuflow/specs/ |

Extraction engine

Universal regex patterns applied to any file:

| Field | Detected from | |-------|--------------| | classes | class, interface, struct, enum, record | | functions | Keyword-prefixed declarations + arrow functions | | dependencies | import, require(), decorators, new ClassName() | | db_tables | SQL FROM/JOIN/INTO, EF DbSet<T>, [Table("…")], _db.TableName | | endpoints | .NET attributes, app.MapGet(), Express router, NestJS, Angular routes | | config_refs | IConfiguration, appsettings, ConnectionStrings:*, process.env.* |

Requirements

• Node.js 18+
• No API keys, no network calls, no AI dependencies

License

MIT — github.com/doquflows/docuflow