rewelo MCPサーバー

Rewelo

Warning: This is experimental software. It comes without support and is not intended for production use.

A prioritization tool that uses the relative weight method to rank stories and tickets by different criteria. Built on SQLite (Node's built-in node:sqlite) for embedded, zero-dependency data storage, exposed through a CLI or as a Docker-based MCP server.

• Website and docs
• Blogpost
• Initial Idea

How It Works

Each ticket is scored on four dimensions using the agile Fibonacci scale (1, 2, 3, 5, 8, 13, 21):

Dimension	Measures
Benefit	Value gained by implementing the story
Penalty	Cost of not implementing the story
Estimate	Resources required for implementation
Risk	Uncertainty or complexity in implementation

From these, the tool calculates priority at runtime:

Value    = Benefit + Penalty
Cost     = Estimate + Risk
Priority = Value / Cost

Higher priority means better return on investment. Relative weights normalize scores across the full backlog or within a tagged subset (e.g. a feature). See calculations.md for the full specification.

Tags

Tickets are organized through a flexible tag system using prefix:value pairs. Tags replace fixed fields for state, feature grouping, and any other classification:

• state:backlog, state:wip, state:done
• feature:checkout, feature:auth
• team:platform, priority:critical

Every tag assignment and removal on a ticket is tracked in an audit log, enabling lead time and cycle time calculations from state: tag transitions.

Revision History

The tool maintains full revision history at two levels:

• Ticket revisions -- snapshots of content (title, description, scores) and tags before each change
• Tag revisions -- snapshots of tag definitions (prefix, value) before renames or edits

This means you can reconstruct the exact state of any ticket at any point in time.

Data Model

The database schema is defined in db/create.sql and documented as DBML in db/model.dbml.

projects
  +-- tickets              (B, P, E, R scores)
  |     +-- ticket_revisions     (content + tag snapshots)
  +-- tags                 (prefix:value pairs)
  |     +-- tag_revisions
  +-- ticket_tags          (current assignments)
  +-- ticket_tag_changes   (audit log: added/removed)
  +-- ticket_relations     (blocks, depends-on, relates-to)
  +-- ticket_deletions     (deleted tickets, for project diff)
  +-- weight_configs       (per-project B/P/E/R weights)

Tech Stack

• SQLite -- embedded database via Node's built-in node:sqlite, no server and no native modules required
• TypeScript -- CLI application and business logic
• Commander.js -- command-line interface
• Docker -- containerized deployment, also usable as an MCP server

CLI Commands

rw project create|list|delete|history|diff
                                    Manage projects, view history and changes
rw ticket  create|list|update|upsert|delete|history
                                    Manage tickets and view revision history
rw tag     create|assign|remove|delete|list|rename|log
                                    Manage tags and view audit log
rw relation create|remove|list|list-all
                                    Ticket relations (blocks, depends-on, relates-to)
rw config  weights                  View/set/reset per-project B/P/E/R weights
rw calc    weights|priority         Relative weights and weighted priority calculations
rw export  csv|json                 Export project data
rw import  csv|json                 Import project data
rw report  summary|group|distribution|health|times|dashboard|event-log
                                    Reporting, HTML dashboard and event stream
rw serve                            Start MCP server (stdio transport)

Global options: --db <path>, --json, --csv, --quiet (and --no-color, accepted for compatibility: rw never prints colour)

Multi-Project Support

The tool is designed around projects as the top-level boundary. Tags, tickets, and all calculations are scoped to a project, so multiple teams or products can use the same instance independently.

Releasing a New Version

Four GitHub Actions workflows in .github/workflows/ cover building, releasing and publishing:

Workflow	Runs on	Does
ci.yml	pushes and pull requests to main	Type check, build, test, and a smoke test of the Docker image
release.yml	pushing a v* tag	Checks the tag matches package.json, tests, and creates the GitHub release with the npm tarball, SBOM, OCI image and changelog; pushes the image to ghcr.io/sebs/rewelo
publish.yml	manual (with a version)	Publishes that tagged version to npm via trusted publishing, with provenance, then lists it in the MCP Registry as io.github.sebs/rewelo
pages.yml	pushes to main	Builds the website from site/ and the docs, and deploys it to GitHub Pages

To release:

1. Bump the version, which commits it (with server.json, the MCP Registry entry, set to the same version) and creates the matching tag:

npm version patch   # 0.1.0 → 0.1.1 (or minor, major)

1. Push the commit and the tag; release.yml builds the release:

git push --follow-tags

1. Publish to npm: run the Publish workflow with the new version (Actions → Publish → Run workflow), or gh workflow run publish.yml -f version=0.1.1. Once npm shows the version, the workflow lists it in the MCP Registry too (mcp-publisher, logged in with the workflow's GitHub OIDC token, so without a secret). Prereleases such as 0.2.0-beta.1 are published under the next dist-tag, and a backport older than the current latest (0.5.2 after 0.6.1) under backport, so npm install rewelo keeps getting the newest stable version.

One-time setup: add sebs/rewelo with workflow publish.yml as a trusted publisher of the rewelo package on npmjs.com, and set Settings → Pages → Source to "GitHub Actions".

To run the released image:

# CLI
docker run --rm -v rw-data:/data ghcr.io/sebs/rewelo:latest project list

# MCP server
docker run --rm -i -v rw-data:/data ghcr.io/sebs/rewelo:latest serve

Use a named volume (rw-data) or a bind mount to persist the database across container restarts. Without one, the database stays inside the container and is lost when it is removed (with --read-only it can't be written at all); rw warns about this on stderr.

MCP Server

The CLI doubles as an MCP server over stdio, letting AI assistants manage projects, tickets, tags, and calculations directly. See mcp.md for client configuration, available tools, and troubleshooting.

Examples

See examples.md for step-by-step scenarios, each as a copy-pasteable Claude Code prompt (MCP) and as rw commands with their output: CSV import, dependency mapping, custom weights, quick wins, tracking work, scripting, and sharing and backing up a backlog.