Registry Review

The Registry Review MCP Server automates the review and validation workflow for carbon credit project registrations. It converts lengthy manual reviews into a guided, auditable process that maps evidence to requirements, performs cross-document checks, and produces structured outputs suitable for both human review and machine processing.

How to use

You interact with the MCP server through a client that supports MCP workflows. Start a new review, let the server discover and classify documents, extract evidence aligned to the checklist, run cross-document validations, and generate a complete review report. The workflow is designed to guide you through recipe-like prompts and to keep an auditable trail of all decisions and findings.

Typical end-to-end usage pattern:

• Initialize a new review session with project metadata
• Run document discovery to classify and index project files
• Execute evidence extraction to map requirements to concrete documents and extract snippets with citations
• Run cross-document validation to verify dates, land tenure consistency, and project IDs
• Generate a report in Markdown and JSON formats with all findings and evidence

How to install

Prerequisites you need on your machine before starting the MCP server are listed here. Ensure you have Python and the UV tool available.

# Prerequisites
# Ensure Python >= 3.10 is installed
# Install the UV package manager (the tool used to run MCP servers)
# This example uses the uv command as shown in the quick start

# Clone or navigate to the project directory
cd regen-registry-review-mcp

# Install dependencies and prepare the environment
uv sync

# Create a configuration file
cp .env.example .env

# Edit .env to set your Anthropic API key and any other options
# Example:
# REGISTRY_REVIEW_ANTHROPIC_API_KEY=sk-...
# REGISTRY_REVIEW_LLM_EXTRACTION_ENABLED=true

# Start the MCP server (stdio configuration)
uv run python -m registry_review_mcp.server

Configuration and runtime notes

You configure the server through a .env file located in the project directory. The following options are commonly adjusted for LLM-based field extraction:

REGISTRY_REVIEW_ANTHROPIC_API_KEY=YOUR_ANTHROPIC_API_KEY
REGISTRY_REVIEW_LLM_EXTRACTION_ENABLED=true
REGISTRY_REVIEW_LLM_MODEL=claude-sonnet-4-20250514
REGISTRY_REVIEW_LLM_CONFIDENCE_THRESHOLD=0.7
REGISTRY_REVIEW_MAX_API_CALLS_PER_SESSION=50
REGISTRY_REVIEW_LOG_LEVEL=INFO

Runtime environment and performance tips

The server supports CPU-only or GPU-accelerated operation. GPU acceleration can significantly speed up PDF marker conversions when processing large document sets. The system uses caching to reduce repeated work, and it provides parallel processing to improve throughput.

# CPU mode: reasonable performance for typical documents
# GPU mode: enable CUDA acceleration if available
# Ensure enough memory and storage are available for model files and caches

Usage examples

A typical end-to-end session uses a sequence of prompts to drive the workflow. Start by initializing the session, then run document discovery, followed by evidence extraction, and finally cross-validation and report generation.

# Stage 1: Initialize a new review
> /initialize Botany Farm 2022-2023, /absolute/path/to/examples/22-23

# Stage 2: Document Discovery
> /document-discovery

# Stage 3: Evidence Extraction
> /evidence-extraction

Available tools

start_review

Quick-start: Create a new review session and automatically discover documents in one step.

create_session

Create a new review session with metadata and checklist loading.

load_session

Load an existing review session for continued work.

list_sessions

List all active review sessions.

delete_session

Delete a session and clean up its data.

discover_documents

Scan and classify project documents across directories.

extract_pdf_text

Extract text and tables from PDF documents with caching.

extract_gis_metadata

Extract GIS shapefile metadata for spatial checks.

extract_evidence

Map checklist requirements to documents and extract evidence snippets.

map_requirement

Map a single requirement to documents with evidence and citations.

cross_validate

Run cross-document validation checks (dates, land tenure, project IDs).

validate_date_alignment

Verify date alignment against the 120-day rule.

validate_land_tenure

Check land tenure consistency with fuzzy name matching.

validate_project_id

Verify project ID patterns and consistency.

generate_review_report

Generate a complete review report in Markdown or JSON.

export_review

Export the review report to a specified location.