@arcbridge/mcp-server
v0.3.0
Published
MCP server for ArcBridge — exposes architecture tools to AI coding agents
Maintainers
Readme
@arcbridge/mcp-server
MCP server for ArcBridge — exposes 34 architecture tools to AI coding agents via the Model Context Protocol.
Install
npm install -g @arcbridge/mcp-serverSetup
Claude Code — add to your project's .mcp.json:
{
"mcpServers": {
"arcbridge": {
"command": "npx",
"args": ["@arcbridge/mcp-server"]
}
}
}Codex CLI — add to ~/.codex/config.toml:
[mcp_servers.arcbridge]
command = "npx"
args = ["-y", "@arcbridge/mcp-server"]Restart your AI agent and approve the MCP server when prompted.
Tools
Lifecycle
| Tool | Description |
|------|-------------|
| arcbridge_init_project | Initialize ArcBridge in a project directory |
| arcbridge_get_project_status | Current phase, task completion, quality summary |
Architecture
| Tool | Description |
|------|-------------|
| arcbridge_get_building_blocks | All blocks with code paths and quality links |
| arcbridge_get_building_block | Deep dive: one block with ADRs, tasks, scenarios |
| arcbridge_get_quality_scenarios | Quality requirements, filterable by category/status |
| arcbridge_get_relevant_adrs | ADRs for a file path or building block |
Planning
| Tool | Description |
|------|-------------|
| arcbridge_get_phase_plan | All phases with tasks and acceptance criteria |
| arcbridge_get_current_tasks | Tasks for the active phase |
| arcbridge_update_task | Mark tasks in-progress, done, blocked, or cancelled |
| arcbridge_create_task | Add a task to any phase |
| arcbridge_delete_task | Remove one or more tasks permanently (batch via task_ids array) |
| arcbridge_create_phase | Add a new phase to the project plan |
| arcbridge_delete_phase | Remove a phase and all its tasks permanently |
Code Intelligence
| Tool | Description |
|------|-------------|
| arcbridge_reindex | Index/re-index TypeScript symbols (incremental) |
| arcbridge_search_symbols | Search by name, kind, file path, or building block |
| arcbridge_get_symbol | Full detail: signature, source, relationships |
| arcbridge_get_dependency_graph | Import/dependency graph for a module |
React & Next.js
| Tool | Description |
|------|-------------|
| arcbridge_get_component_graph | Component hierarchy with props, state, context flow |
| arcbridge_get_route_map | Next.js App Router routes with layouts and middleware |
| arcbridge_get_boundary_analysis | Server/client boundary analysis |
Architecture Bridge
| Tool | Description |
|------|-------------|
| arcbridge_check_drift | Detect drift between docs and code |
| arcbridge_get_guidance | Context-aware guidance for a file or block |
| arcbridge_get_open_questions | Unresolved architectural questions and risks |
| arcbridge_propose_arc42_update | Generate arc42 update proposals from code changes |
| arcbridge_get_practice_review | 5-dimension review: architecture, security, testing, docs, complexity |
| arcbridge_update_arc42_section | Read or update any arc42 markdown section (frontmatter preserved) |
Roles & Sync
| Tool | Description |
|------|-------------|
| arcbridge_complete_phase | Validate phase gates and transition to next phase |
| arcbridge_activate_role | Load agent role with tools and pre-loaded context |
| arcbridge_verify_scenarios | Run linked tests for quality scenarios |
| arcbridge_update_scenario_status | Manually update scenario status and link test files |
| arcbridge_run_role_check | Run a role's quality checks against code |
Metrics
| Tool | Description |
|------|-------------|
| arcbridge_record_activity | Record agent activity — model, tokens, cost, duration, quality snapshot |
| arcbridge_get_metrics | Query and aggregate activity by model, task, phase, tool, or day |
| arcbridge_export_metrics | Export metrics to JSON, CSV, or Markdown for git commits |
How It Works
The server communicates over stdio using the MCP protocol. Each tool call receives a target_dir parameter pointing to an ArcBridge-initialized project. The server manages a SQLite database (.arcbridge/index.db) that caches architecture docs, indexed symbols, and planning state.
Data flow: YAML/markdown (source of truth) -> SQLite (query cache) -> MCP tools (agent interface)
All mutations (task updates, phase transitions, scenario results) write back to both the database and the source YAML files, so the docs stay in sync.
