fileman
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@fileman organize my Downloads folder"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
fileman is a local-first file organizer. Drop a messy folder. Get a manifest, a plan, and a one-click apply. Your originals never move until you say so.
┌──────────────────────────────────────┐
│ LOCAL-FIRST ████████ 100% │
│ SOURCE-TRACEABLE ████████ 100% │
│ TYPING REQUIRED ░░░░░░░░ 0% │
│ VIBES ████████ ZERO │
│ FILLER│
└──────────────────────────────────────┘AI-assisted file triage. Manifest-first, never destructive.
What You Get
Surface | What |
| Scan, plan, manifest. Apply only when you nod. |
| Browse the plan. Approve, edit, or send back for another pass. |
| Same engine, exposed as an MCP server for any agent. |
| Plans are diffable. Plans are reviewable. Plans are receipts. |
| Drop into Claude/Codex/OpenClaw. Hand it a folder. Get a library. |
Local-first by default. No silent telemetry. No cloud round-trip. Your data stays on your machine until you explicitly ship it somewhere.
Related MCP server: Organizer MCP Server
Install
git clone https://github.com/xiaojiou176-open/fileman.git
cd fileman
# follow the per-stack quickstart in INSTALL.md or docs/Three commands. No curl | sh. No login. Read what you run.
Install break? Open your favorite agent and say "Read AGENTS.md and bootstrap fileman for me." Agent fix own brain. Long version: docs/.
How It Work
The repo is seven layers — exactly the seven commits in git log. New work goes in as small named PRs. No 50-file mystery commits.
Layer | What |
| License, governance, hygiene gates, CI scaffolding. |
| The primary engine. The reason fileman exists. |
| Packages, adapters, services, plugins. The second floor. |
| Schemas, configs, public boundaries. Other code talks here. |
| Receipts. Everything in this layer must run. |
| Scripts, infra, CI helpers, build glue. |
| Public docs surface. The pretty face. |
git log reads like a building floor plan. Look once, know the whole shape.
Ecosystem
fileman lives in the yard family: seven yards. one philosophy: structured input, structured output, structured proof.
Repo | What |
model & agent runtime switch board | |
your AI clicks, your browser obeys | |
your Apple Notes never really die | |
let prices fight, you sit and watch | |
docs site in, markdown out, no scraping by hand | |
fileman (you here) | messy folders in, organized library out |
every claim ships with its receipt |
Cross-family taste: BeamMe · BrewMe · OpenVibeCoding · proofyard.
Star This Repo
If fileman saves you a click, an hour, or a headache — star costs zero. Fair trade. ⭐
Also by Yifeng[Terry] Yu
switchyard — model & agent runtime switch board
browserclickyard — your AI clicks, your browser obeys
BeamMe — beam your agent config to any planet
BrewMe — wake up, news already brewed
OpenVibeCoding — AI codes overnight, you ship in the morning
License
MIT — small print, big freedom.
Available Tools
16 toolsanalyze.createC
Create an analyze job without mutating files. Review still happens later.
Inputs: input_directory (required), model, offline, workers, +7 more. Returns: object
apply.previewC
Create an apply job in dry-run mode only.
Inputs: output_root, manifest_path, analyze_job_id. Returns: object
inbox.analyzeC
Create an analyze job from one watch source and optional discovered batch id.
Inputs: watch_source_id (required), model, offline, workers, +6 more. Returns: object
inbox.scanARead-onlyIdempotent
Run inbox discovery only. This does not auto-start analyze or apply.
Inputs: no named parameters. Returns: object
jobs.getCRead-onlyIdempotent
Get one Fileman job by id.
Inputs: job_id (required). Returns: object
jobs.listARead-onlyIdempotent
List current Fileman jobs from the local workspace job store.
Inputs: no named parameters. Returns: object
manifest.batch_patchB
Patch multiple manifest rows through the overlay only.
Inputs: job_id (required), operations (required). Returns: object
manifest.getBRead-onlyIdempotent
Get the manifest view for one job. Defaults to the resolved review view.
Inputs: job_id (required), limit, resolved. Returns: object
manifest.patch_rowB
Patch one manifest row through the overlay only.
Inputs: patch (required), job_id (required), row_id (required). Returns: object
report.getARead-onlyIdempotent
Get one persisted report plus the review bridge metadata.
Inputs: job_id (required). Returns: object
review_queue.getCRead-onlyIdempotent
Get the review queue with deterministic copilot hints and collection slices.
Inputs: job_id (required), limit. Returns: object
review_rule.applyC
Apply a review rule to the overlay only. This does not execute file changes.
Inputs: job_id (required), rule, rule_id. Returns: object
review_rule.previewARead-onlyIdempotent
Preview a review rule against the current review queue without changing anything.
Inputs: job_id (required), rule, rule_id. Returns: object
runtime.settings.getARead-onlyIdempotent
Read the current local runtime settings.
Inputs: no named parameters. Returns: object
strategy_packs.listARead-onlyIdempotent
List repo-shipped strategy packs and the active local default.
Inputs: no named parameters. Returns: object
watch_sources.listARead-onlyIdempotent
List local-first watch sources and their linked strategy packs.
Inputs: no named parameters. Returns: object
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 16 tools
Each tool targets a distinct action (list, get, create, preview, apply, scan, etc.) across different domains (jobs, manifest, review rules, inbox). Overlap is minimal, with only 'analyze.create' and 'inbox.analyze' slightly related but clearly differentiated by source.
All tools follow a consistent `domain.action` or `domain.subdomain.action` pattern using lower_snake_case. Verbs like list, get, create, preview, apply, scan are used uniformly, making the naming predictable.
With 16 tools, the set is well-scoped for a file management system covering jobs, manifests, review rules, inbox processing, settings, and reports. No tool feels redundant or missing in terms of count.
The tool surface covers core workflows (analyze, review, manifest patching, inbox scanning) but lacks execution tools for applying changes (only preview exists), job cancellation, or manifest creation. Some lifecycle gaps exist.
Maintenance
Related MCP Connectors
Your personal library, readable by AI agents: durable snapshots, hybrid search, 46 MCP tools.
Your own computer's files from ChatGPT or Claude: folder rules, confirmations and undo.
Search photos and video moments by visual meaning, people, dates, and folders across local libraries, Apple Photos, and OneDrive. Manage collections, groups, favorites, people, and duplicate review. Synced metadata remains searchable while the desktop is offline; local and Apple Photos media access needs the owning device online with remote access enabled. OneDrive media can be accessed directly from the cloud. OAuth or Bearer API keys. Paid Cloud Access required; hosted visual search uses prepaid credits. Text-in-image search is semantic matching, not guaranteed exact OCR. Setup: https://organizewith.ai/en/guides/connect-ai-assistants
Your org's AI agents, tasks, runs, search, and brain files as MCP tools and resources.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceEnables LLMs to organize and manage Windows file systems with intelligent file analysis, automated grouping, renaming with date prefixes, and safe operations through dry-run mode and sandbox restrictions.1MIT
- AlicenseNot gradedqualityDmaintenanceEnables Claude to intelligently organize files by scanning folders, detecting duplicates, suggesting meaningful names, and moving files into logical folders.MIT
- FlicenseNot gradedqualityDmaintenanceLets AI assistants analyze and organize local directories by file type, with dry-run support and safe sorting.-
- AlicenseNot gradedqualityCmaintenanceA local-first, open-source harness for organizing your files with an AI agent.2MIT