Which integrations are available for this server?

Utilized for local data analytics and entity resolution within the building permit data processing pipeline. Provides tools for searching and analyzing millions of building permit records, entity relationships, and plan review routing data stored in a PostgreSQL database.

How do I use SF Permits MCP Server?

1. Click on "Install Server". 2. Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state. 3. In the chat, type @ followed by the MCP server name and your instructions, e.g., "@SF Permits MCP Server Show me recent building permits in the Mission District" That's it! The server will respond to your query, and you can continue using it as needed. Here is a step-by-step guide with screenshots.

SF Permits MCP Server

MCP server + web application for San Francisco building permit data, entity network analysis, AI-powered permit guidance, and AI vision plan analysis. Built with FastMCP, Flask + HTMX, and deployed on Railway.

Live: https://sfpermits-ai-production.up.railway.app

Tools (21 MCP Tools)

Phase	Tool	Description
1 — SODA API	`search_permits`	Search building permits by neighborhood, type, status, cost, date, address
	`get_permit_details`	Full details for a specific permit by number
	`permit_stats`	Aggregate statistics grouped by neighborhood, type, status, month, year
	`search_businesses`	Search registered business locations
	`property_lookup`	Property assessments by address or block/lot
	`search_complaints`	DBI complaint records
	`search_violations`	Notices of violation
	`search_inspections`	Building inspection records
2 — Entity/Network	`search_entity`	Find entities by name across 1M+ resolved records
	`entity_network`	N-hop network traversal for entity relationships
	`network_anomalies`	Anomaly detection across entity networks
2.75 — Knowledge	`predict_permits`	Predict required permits via decision tree walk
	`estimate_timeline`	Timeline estimates from historical percentiles
	`estimate_fees`	Fee calculation from structured fee tables
	`required_documents`	Document checklist assembly
	`revision_risk`	Revision probability from cost/code analysis
3.5 — Facilitation	`recommend_consultants`	Land use consultant recommendations
	`permit_lookup`	Quick permit lookup by number, address, or parcel — exact matching, historical lot discovery, parcel-level merge
4 — Vision	`analyze_plans`	AI vision analysis of architectural drawings
	`validate_plans`	EPR compliance checking via Claude Vision
5 — Addenda	`search_addenda`	Search 3.9M+ plan review routing records by permit, station, reviewer, date

Data Sources

All data from DataSF via the Socrata SODA API. 22 datasets cataloged, 13.3M records:

Permits: Building (1.3M), Plumbing (513K), Electrical (344K), Boiler (152K), Street-Use (1.2M)
Contacts: Building (1M), Electrical (340K), Plumbing (503K) — resolved into 1M+ entities
Violations: Inspections (671K), Complaints (326K), Notices of Violation (509K)
Enrichment: Business Locations (354K), Property Tax Rolls (3.7M), Development Pipeline, Housing Production

See datasets/CATALOG.md for the full catalog.

Architecture

Users (browser)
    |
    v
Flask + HTMX Web UI (Railway)  <-- https://sfpermits-ai-production.up.railway.app
    |
    |--- PostgreSQL (pgvector-db) -- users, auth, RAG embeddings, permit tracking
    |--- SODA API (data.sfgov.org) -- live permit queries
    |--- Claude Vision API ---------- plan analysis
    |
Claude (claude.ai / Claude Code)
    |
    v
FastMCP Server — 21 tools
    |--- Phase 1 (8 tools) -------> SODA API (live HTTP)
    |--- Phase 2 (3 tools) -------> PostgreSQL (entities, relationships)
    |--- Phase 2.75 (5 tools) ----> Knowledge Base (39 tier1 JSON files)
    |--- Phase 3.5 (2 tools) -----> PostgreSQL + Knowledge Base
    |--- Phase 4 (2 tools) -------> Claude Vision API
    |--- Phase 5 (1 tool) -------> PostgreSQL (addenda routing, 3.9M rows)

Key Numbers

Metric	Value
MCP tools	21
SODA datasets	23 (17.2M records)
Entities	1M+ (resolved from 1.8M contacts)
Relationship edges	576K
Addenda routing records	3.9M
Knowledge base	39 tier1 JSON files, ~86 semantic concepts
RAG chunks	3,682 (pgvector embeddings)
Tests	1,075+
PostgreSQL tables	21

Setup

git clone https://github.com/tbrennem-source/sf-permits-mcp.git
cd sf-permits-mcp
pip install -e ".[dev]"

# MCP server
python -m src.server

# Web UI (local)
python -m web.app

# Tests
pytest tests/ -v

Project Phases

Phase 1: MCP server + SODA API tools + dataset catalog + benchmarks
Phase 2: DuckDB local analytics, entity resolution, co-occurrence graph
Phase 2.75: Knowledge base, decision tree, permit guidance tools
Phase 3: Web UI (Flask + HTMX), auth, morning briefs, feedback
Phase 3.5: Railway deployment, PostgreSQL migration, regulatory watch, consultant recommendations
Phase 4 (partial): AI Vision plan analysis, EPR compliance checking
Phase 4 (remaining): RAG activation, nightly refresh

Documentation

Document	Purpose
`CLAUDE.md`	Primary reference — project structure, Railway infra, deploy instructions
`docs/ARCHITECTURE.md`	Data flow, schemas, module details
`docs/DECISIONS.md`	Architecture decision log
`docs/BACKUPS.md`	Backup strategy and recovery playbook
`CHANGELOG.md`	Session-by-session build log
`data/knowledge/SOURCES.md`	Knowledge base inventory