Claude CI Dashboard

CI dashboard for managing Claude Code workflows on Forgejo. Three interfaces: MCP server, web GUI, and TUI.

Components

File	Description
mcp_server.py	FastMCP server — tools for Claude Code to manage issues, runs, PRs
web.py	Flask web GUI — kanban board with issue management
dashboard.py	Textual TUI — terminal dashboard
forgejo_api.py	Forgejo API client (urllib, no deps)
state.py	State machine: new -> clarifying -> planning -> implementing -> reviewing -> merged

Setup

pip install flask fastmcp textual rich

Environment Variables

Variable	Default	Description
FORGEJO_URL	https://silo.retromonkey.com.au	Forgejo instance URL (LAN fallback: http://192.168.0.33:3200)
FORGEJO_TOKEN		Forgejo API token
FORGEJO_REPO	aaron/claude-ci-demo	Target repo (owner/repo)

MCP Server Setup

Add to Claude Code (user scope)

claude mcp add claude-ci -s user -- python3 /path/to/claude_ci_mcp.py \
  -e FORGEJO_URL=https://silo.retromonkey.com.au \
  -e FORGEJO_TOKEN=<your-token> \
  -e FORGEJO_REPO=aaron/claude-ci-demo

Available Scopes

Scope	Flag	Persist across projects?	Use case
local	-s local (default)	No	One-off testing
user	-s user	Yes	Personal setup, all projects
project	-s project	Per project (.mcp.json)	Team-shared config

Managing MCP Servers

# List all servers (with health check)
claude mcp list

# Get details for one server
claude mcp get claude-ci

# Remove a server
claude mcp remove -s user claude-ci

# Reset all project-scoped approvals
claude mcp reset-project-choices

# Add SSE server (e.g. searchmcp)
claude mcp add searchmcp -s user --transport sse --url http://192.168.0.33:8013/sse

Blocking Unwanted Servers

Add to ~/.claude/settings.json:

{
  "deniedMcpServers": ["unwanted-server-name"]
}

Note: Some servers may be injected by the API provider (e.g. Z.ai) and cannot be blocked this way.

MCP Tools

The server exposes the following tools to Claude Code:

Tool	Description
ci_dashboard	Full CI dashboard summary: issues, runs, runner status
ci_create_issue	Create a new issue (title, body)
ci_get_issue	Get issue details including comments
ci_list_issues	List issues with state labels (open/closed/all)
ci_reply	Comment on an issue
ci_approve	Approve an issue — adds approved comment, sets awaiting-approval label
ci_list_runs	List recent workflow runs
ci_runner_status	Check runner health and status
ci_list_pulls	List pull requests (open/closed/all)
ci_get_pull	Get pull request details

Running

Web GUI (port 8100)

python web.py --port 8100

MCP Server

python mcp_server.py

TUI

python dashboard.py --refresh 10

CI State Machine

NEW → CLARIFYING → PLANNING → IMPLEMENTING → REVIEWING → MERGED

Issues flow through states via labels:

• awaiting-clarification → clarifying
• awaiting-approval → planning
• awaiting-review → reviewing

Web GUI Features

• Kanban board with columns per state
• Issue detail panel with comments
• Create/reply/approve actions
• Workflow runs sidebar with status
• State machine visualization
• Runner health indicator
• Auto-refresh every 30s
• Keyboard shortcuts: R refresh, N new issue, Esc close