remembril-mcp 0.2.0

MCP server for Remembril - change impact tracking, ripple effects, Graphify bridge, and autonomous coding loop

Remembril MCP Server

MCP (Model Context Protocol) server for Remembril — change impact tracking, autonomous coding loops, Graphify knowledge-graph bridge, and documentation automation for AI coding assistants.

57 tools across 9 categories, exposed over stdio and streamable HTTP. Three output tiers: MCP Apps widgets (Claude Desktop / ChatGPT), A2UI declarative JSON (Google Gemini), and plain-text markdown (Claude Code CLI and any MCP host).

What is Remembril?

Remembril tracks changes through your entire documentation hierarchy: Requirements, PRFAQ, HLD, LLD, Tests, Compliance, and Runbooks. When something changes, ripple effects propagate automatically. The autonomous coding loop can then implement those changes via Claude Code with an AI review committee as quality gate. The Graphify bridge overlays code-level symbols from safishamsi/graphify into the same knowledge graph, so requirements and actual code live in one connected graph.

Installation

pip install remembril-mcp

# Optional: include the HTTP transport for Claude Desktop / ChatGPT / Gemini
pip install 'remembril-mcp[http]'

Quick Start

1. Setup for Claude Code

Run the setup script (if you have the repo):

cd mcp-server
./setup-claude-code.sh

Or configure manually in ~/.claude/settings.json:

{
  "mcpServers": {
    "remembril": {
      "command": "remembril-mcp",
      "args": ["serve"],
      "env": {
        "REMEMBRIL_URL": "http://localhost:7070",
        "REMEMBRIL_TOKEN": "your-token-here"
      }
    }
  }
}

2. Authenticate

remembril-mcp login --url http://localhost:7070

3. Create a project and configure it

From Claude Code, just say:

"Create a Remembril project for this repo and set up the coding loop"

The AI will call remembril_project_create and remembril_project_configure with your repo path, test command, etc.

Or via CLI:

remembril-mcp projects                    # List projects
remembril-mcp set-project YOUR_PROJECT_ID # Set default

Available MCP Tools

Project Management

Tool	Description
remembril_projects	List all projects
remembril_project_create	Create a new project
remembril_project_configure	Configure repo path, test command, agent review, coding loop
remembril_project_settings	View current pipeline settings

Issues & Ripples

Tool	Description
remembril_issue_create	Create a new issue
remembril_issue_create_smart	Create issue with auto-severity
remembril_issue_list	List issues with filters
remembril_ripples_generate	Generate ripple effects for an issue
remembril_ripples_queue	Get pending ripples for review
remembril_ripple_approve	Approve a ripple
remembril_ripple_reject	Reject a ripple with reason
remembril_ripple_update	Modify a ripple's changes
remembril_ripple_apply	Apply an approved ripple to its document
remembril_ripple_diff	View the diff a ripple would produce
remembril_ripple_blind_review	Get ripples with AI reasoning hidden (blind validation)

Pipeline & Documents

Tool	Description
remembril_pipeline_run	Run the full auto-pipeline (issues -> ripples -> docs)
remembril_pipeline_status	Check pending work for a project
remembril_pipeline_generate_docs	Export documents to markdown
remembril_docs	List project documents
remembril_doc_read	Read a specific document
remembril_doc_traceability	View cross-document traceability matrix

Autonomous Coding Loop

Tool	Description
remembril_loop_start	Start a coding loop (Issue -> Code -> Test -> Review -> Merge)
remembril_loop_status	Check loop progress, waves, and droplets
remembril_loop_stop	Stop a running loop

Pull-Based Work Loop (v0.2.0)

Agents pull atomic droplets over MCP instead of being driven by a subprocess. Respects depends_on_id chains — blocked droplets stay hidden until their dependency completes.

Tool	Description
remembril_loop_next_task	Get the next available droplet for the calling agent
remembril_loop_claim_task	Claim a droplet so no other worker picks it up
remembril_loop_checkpoint	Periodic progress update while working
remembril_loop_complete_task	Mark a droplet complete, unblocks downstream

Graphify Bridge (v0.2.0)

Integrates with safishamsi/graphify (PyPI graphifyy) — a tree-sitter + Leiden knowledge graph builder. Diffs between snapshots become issue-shaped ripple payloads for the change-impact pipeline.

Tool	Description
remembril_graphify_import	Parse a Graphify graph.json and map it into Remembril's knowledge graph
remembril_graphify_diff	Diff two snapshots, emit issue-shaped ripple payloads for added/removed nodes and edges
remembril_graphify_sync	Rebuild the Graphify graph for a project dir (requires graphifyy on the backend)

Agent Review (Auror Committee)

Tool	Description
remembril_review_run	Run an AI review committee on content
remembril_review_status	Check review status
remembril_review_list	List reviews for a project
remembril_review_approve_ripples	Approve ripples from a review
remembril_debate_create	Start a debate between agents
remembril_debate_status	Check debate progress
remembril_watcher_alerts	View anomaly alerts from the watcher
remembril_alert_handle	Handle a watcher alert
remembril_personas_list	List available review personas

Code Intelligence

Tool	Description
remembril_scan	Index a codebase for symbol tracking
remembril_impact	Analyze file change impact
remembril_trace	Trace requirement to code
remembril_symbols	Search code symbols

Orchestration

Tool	Description
remembril_orchestrate_status	Check pending work
remembril_orchestrate_start	Start an orchestration session
remembril_orchestrate_next	Get next task in session
remembril_orchestrate_complete	Complete a task
remembril_orchestrate_sessions	List sessions
remembril_orchestrate_generate_all	Generate all pending ripples

CLI Commands

# Authentication
remembril-mcp login              # Login via browser
remembril-mcp login -t TOKEN     # Login with token
remembril-mcp logout             # Remove credentials
remembril-mcp status             # Check connection

# Projects
remembril-mcp projects           # List projects
remembril-mcp set-project ID     # Set default project

# Server
remembril-mcp serve              # Start MCP server (stdio)

# Issues
remembril-mcp issue "Title" -d "Description"
remembril-mcp issue "Title" --auto-severity

# Ripple Workflow
remembril-mcp run 123            # Generate ripples for issue
remembril-mcp review 123         # Interactive review
remembril-mcp review 123 --blind # Blind validation
remembril-mcp resume 123         # Resume reviewing
remembril-mcp logs               # View history

Typical Workflow

1. Create project           -> remembril_project_create
2. Configure repo + tests   -> remembril_project_configure
3. Create issues             -> remembril_issue_create
4. Run pipeline              -> remembril_pipeline_run
   (generates ripples across Requirements, PRFAQ, HLD, LLD, Tests, Compliance, Runbook)
5. Start coding loop         -> remembril_loop_start
   (Claude Code implements changes, tests run, Auror committee reviews, auto-merge)
6. Check progress            -> remembril_loop_status

Environment Variables

Variable	Description
REMEMBRIL_URL	API URL (default: https://www.remembril.com)
REMEMBRIL_TOKEN	Authentication token
REMEMBRIL_PROJECT_ID	Default project ID

HTTP Transport (v0.2.0)

Run the server over streamable HTTP instead of stdio — required for Claude Desktop, ChatGPT, and Google Gemini hosts that don't support stdio subprocesses.

pip install 'remembril-mcp[http]'
remembril-mcp serve --http --port 8080

Exposes:

Path	Purpose
POST /mcp	StreamableHTTP transport (stateless sessions per request)
GET /sse, POST /messages/	Server-Sent Events transport (legacy)
GET /health	Health check

Claude Desktop / ChatGPT config

{
  "mcpServers": {
    "remembril": {
      "type": "http",
      "url": "https://mcp.remembril.com/mcp"
    }
  }
}

Output Tiers

Every tool supports three output tiers so any MCP host gets the best available rendering:

1. MCP Apps — HTML widget bundles (Claude Desktop, ChatGPT). Five widgets: ripple-review, traceability, pipeline-dashboard, design-preview, agent-review.
2. A2UI — declarative JSON (Google Gemini). Includes Cards, DataTable, ProgressBar, DesignPreview, AgentReview, WatcherAlerts.
3. Text — markdown fallback (Claude Code CLI, any MCP host).

Configuration

Credentials stored in ~/.remembril/config.json (600 permissions).

License

MIT