Public beta

Every civic dataset. One API.

REST for per-property dossiers. SQL for analytics over 2,500+ datasets. AI for natural-language investigations. Time-travel built in. $0 egress forever.

Live2284datasets854.9Mrows indexed$0egress forever

Get API key Interactive reference Try SQL in-browser MCP for AI assistants

Three surfaces

Pick the abstraction that fits

Same data. Three ways to query it. All metered against the same credit balance.

REST

1–10 credits

/api/v1/property/* /api/v1/owner/* /api/v1/area/*

Curated per-entity dossiers. Pre-joined zoning + violations + permits + ownership per property. Fixed cost per call. Good when you know exactly what you're asking.

SQL

~1 credit / 10 MB scanned

/api/v1/data/query

DuckLake SQL over 2,500+ civic datasets via R2 parquet. Time-travel (`as_of`), Parquet output (`format=parquet`), and federated joins against Postgres all supported. Credit cost scales with bytes scanned.

AI

5–50 credits

/api/v1/answers

Natural-language Q&A over the same data. The system routes between tools and the knowledge graph, cites every source, and explains its reasoning. Good for investigators + exploratory research.

Quick start

Your first call in 60 seconds

Sign up at the developer portal — first 5,000 credits on us, no credit card.
Generate an API key and save it as MUNIMIND_API_KEY.
Call any endpoint below.

curl https://api.munimind.com/api/v1/property/1000010010/summary \
  -H "X-API-Key: $MUNIMIND_API_KEY"

SQL

Full DuckLake SQL over R2

The /api/v1/data/query endpoint runs sandboxed DuckDB over 2,500+ datasets stored as parquet on Cloudflare R2. Joins, aggregations, window functions, CTEs — all supported. Credit cost = ~1 per 10 MB scanned.

Aggregate across a city

curl -X POST https://api.munimind.com/api/v1/data/query \
  -H "X-API-Key: $MUNIMIND_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "sql": "SELECT borough, count(*) AS n FROM lake.main.\"nyc__hpd_violations\" WHERE inspection_date >= '\''2025-01-01'\'' GROUP BY borough ORDER BY n DESC"
  }'

Time-travel (unique to DuckLake)

Every dataset we ingest is snapshotted. Pass as_of to query the state at any point in history — legal-grade for diligence and audit.

# DuckLake time-travel — 'what did this look like on a past date?'
curl -X POST https://api.munimind.com/api/v1/data/query \
  -H "X-API-Key: $MUNIMIND_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "sql": "SELECT count(*) FROM lake.main.\"nyc__dob_violations\"",
    "as_of": "2026-01-15T00:00:00"
  }'

Parquet output

Pass format: "parquet" for Arrow-compatible binary output. Pandas / Polars / DuckDB read it natively with zero copies.

# Return results as Apache Parquet for pandas/polars/DuckDB clients
curl -X POST https://api.munimind.com/api/v1/data/query \
  -H "X-API-Key: $MUNIMIND_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"sql":"SELECT * FROM lake.main.\"nyc__taxi_zones\"","format":"parquet"}' \
  --output taxi_zones.parquet

Safe retries (Idempotency-Key)

Include Idempotency-Key on any POST — same key + body within 24 h returns the cached response, with Idempotency-Replay: true. You never pay for a retry twice.

# Safe retries — same key + same body returns the cached response
# (within 24h). Per IETF idempotency-key-header draft.
curl -X POST https://api.munimind.com/api/v1/data/query \
  -H "X-API-Key: $MUNIMIND_API_KEY" \
  -H "Idempotency-Key: $(uuidgen)" \
  -H "Content-Type: application/json" \
  -d '{"sql":"SELECT * FROM lake.main.\"nyc__taxi_zones\" LIMIT 10"}'

Prefer REST-style filters? We have those too.

# Filter rows PostgREST-style, no SQL needed
curl 'https://api.munimind.com/api/v1/data/nyc.dob_violations/rows?where=bbl.eq.1000010010&limit=100' \
  -H "X-API-Key: $MUNIMIND_API_KEY"

Arrow streaming (Accept: arrow.stream)

Content-negotiate Arrow IPC stream via Accept: application/vnd.apache.arrow.stream for 10-100x smaller payloads. ADBC/pandas/polars/DuckDB read it zero-copy. Same credit cost as JSON.

# Request Apache Arrow IPC stream — 10-100x smaller than JSON,
# zero-copy into pandas / polars / DuckDB.
curl -X POST https://api.munimind.com/api/v1/data/query \
  -H "X-API-Key: $MUNIMIND_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Accept: application/vnd.apache.arrow.stream" \
  -d '{"sql":"SELECT * FROM lake.main.\"nyc__taxi_zones\""}' \
  --output taxi_zones.arrow

# Python — native Arrow reader:
python3 -c "
import pyarrow.ipc as ipc
with open('taxi_zones.arrow','rb') as f:
    reader = ipc.open_stream(f)
    df = reader.read_all().to_pandas()
print(df.head())"

Geospatial — /area/tilequery

Point + radius query returns everything nearby as a GeoJSON FeatureCollection. Parcels come from PostGIS; permits + violations attach via BBL. Drop straight into Mapbox / Deck.gl / Leaflet. 3 credits / call, radius capped at 5 km.

# Point-radius GeoJSON — Mapbox-style
curl "https://api.munimind.com/api/v1/area/tilequery?lng=-73.986&lat=40.736&radius_m=250&city=nyc&types=parcels,permits,violations&limit=200" \
  -H "X-API-Key: $MUNIMIND_API_KEY"
# Returns { type: "FeatureCollection", features: [...] } — drop straight
# into Mapbox, Deck.gl, Leaflet, etc. 3 credits / call, 5 km radius cap.

Ask in plain English

Our agent routes between tools, cites every source, and explains its reasoning. Good for investigators doing exploratory work.

curl -X POST https://api.munimind.com/api/v1/answers \
  -H "X-API-Key: $MUNIMIND_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"question":"Who are the top 3 owners on the Lower East Side by HPD violations last year?"}'

Auth + billing

Keys, credits, rate limits

X-API-Key

Every request sends X-API-Key: mm_live_.... Keys scope to a user, tier, and rate limit. Hash-based lookup — we never store keys in plain text. Canary tokens auto-detect compromise.

RateLimit headers

Every response carries IETF-standard RateLimit + RateLimit-Policy + de-facto X-RateLimit-* headers. You know exactly how much budget you have left.

Credit system

Everything costs credits. Credits roll over monthly. Use our live usage page for per-endpoint spend, top-scanning queries, and projected run-rate.

Property lookup (REST)1 credit

Property intelligence dossier5 credits

Owner portfolio10 credits

SQL scan of 50 MB~5 credits

SQL scan of 500 MB~50 credits

AI answer (simple)5 credits

AI answer (deep investigation)25–50 credits

Errors

401 — missing or invalid key.
402 — insufficient credits. Top up at /pricing.
422 — Idempotency-Key already used with different body.
429 — rate limit exceeded. Check Retry-After header.
400 — sandbox rejection (DDL, multi-statement, etc.).

Pricing

Simple credit-based pricing

First 5,000 credits free. No credit card. Upgrade when you need more throughput or overage pricing.

Free

5,000 credits on signup

10 req/min

5,000 free credits to start
All 2,500+ datasets
Catalog + previews always free
Full SDK + MCP server access

Get API key

Pro

$79/ mo

5,000 included / mo

120 req/min

5,000 included + $0.005 / extra call
Priority routing
Email support
All features including time-travel

Upgrade

Enterprise

$299/ mo

50,000 included / mo

600 req/min

50,000 included + $0.003 / extra call
Dedicated Slack
SLA + uptime guarantees
Volume pricing

Contact sales

Dataset catalog, schemas, and first-100-row previews are always free + unauthenticated. Try any dataset before signing up.

Live catalog

Featured datasets

View all 2284 →

nyc.2019_high_volume_fhv_trip_records

234.1M rows

Civic dataset from NYC Open Data, migrated 2026-04-19.

nyc.2015_yellow_taxi_trip_data

146.1M rows

Civic dataset from NYC Open Data, migrated 2026-04-19.

nyc.2017_yellow_taxi_trip_data

113.5M rows

Civic dataset from NYC Open Data, migrated 2026-04-19.

nyc.2019_yellow_taxi_trip_data

84.2M rows

Civic dataset from NYC Open Data, migrated 2026-04-19.

nyc.2015_for_hire_vehicle_trip_data

63.9M rows

Civic dataset from NYC Open Data

nyc.2020_high_volume_fhv_trip_records

31.4M rows

Civic dataset from NYC Open Data

Python SDK

Auto-generated from our OpenAPI spec. All 57 endpoints fully typed. Regenerates on every API change.

pip install munimind

MCP for AI assistants

Plug MuniMind into Claude Desktop / Code / Cursor. 7 tools + 5 prompts + 4 resources.

MCP setup →

Ready to build?

Get API key Interactive reference

MuniMind

Public beta

Every civic dataset. One API.

REST for per-property dossiers. SQL for analytics over 2,500+ datasets. AI for natural-language investigations. Time-travel built in. $0 egress forever.

Live2284datasets854.9Mrows indexed$0egress forever

Get API key Interactive reference Try SQL in-browser MCP for AI assistants

Three surfaces

Pick the abstraction that fits

Same data. Three ways to query it. All metered against the same credit balance.

REST

1–10 credits

/api/v1/property/* /api/v1/owner/* /api/v1/area/*

Curated per-entity dossiers. Pre-joined zoning + violations + permits + ownership per property. Fixed cost per call. Good when you know exactly what you're asking.

SQL

~1 credit / 10 MB scanned

/api/v1/data/query

AI

5–50 credits

/api/v1/answers

Natural-language Q&A over the same data. The system routes between tools and the knowledge graph, cites every source, and explains its reasoning. Good for investigators + exploratory research.

Quick start

Your first call in 60 seconds

Sign up at the developer portal — first 5,000 credits on us, no credit card.
Generate an API key and save it as MUNIMIND_API_KEY.
Call any endpoint below.

curl https://api.munimind.com/api/v1/property/1000010010/summary \
  -H "X-API-Key: $MUNIMIND_API_KEY"

SQL

Full DuckLake SQL over R2

Aggregate across a city

curl -X POST https://api.munimind.com/api/v1/data/query \
  -H "X-API-Key: $MUNIMIND_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "sql": "SELECT borough, count(*) AS n FROM lake.main.\"nyc__hpd_violations\" WHERE inspection_date >= '\''2025-01-01'\'' GROUP BY borough ORDER BY n DESC"
  }'

Time-travel (unique to DuckLake)

Every dataset we ingest is snapshotted. Pass as_of to query the state at any point in history — legal-grade for diligence and audit.

# DuckLake time-travel — 'what did this look like on a past date?'
curl -X POST https://api.munimind.com/api/v1/data/query \
  -H "X-API-Key: $MUNIMIND_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "sql": "SELECT count(*) FROM lake.main.\"nyc__dob_violations\"",
    "as_of": "2026-01-15T00:00:00"
  }'

Parquet output

Pass format: "parquet" for Arrow-compatible binary output. Pandas / Polars / DuckDB read it natively with zero copies.

# Return results as Apache Parquet for pandas/polars/DuckDB clients
curl -X POST https://api.munimind.com/api/v1/data/query \
  -H "X-API-Key: $MUNIMIND_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"sql":"SELECT * FROM lake.main.\"nyc__taxi_zones\"","format":"parquet"}' \
  --output taxi_zones.parquet

Safe retries (Idempotency-Key)

Include Idempotency-Key on any POST — same key + body within 24 h returns the cached response, with Idempotency-Replay: true. You never pay for a retry twice.

# Safe retries — same key + same body returns the cached response
# (within 24h). Per IETF idempotency-key-header draft.
curl -X POST https://api.munimind.com/api/v1/data/query \
  -H "X-API-Key: $MUNIMIND_API_KEY" \
  -H "Idempotency-Key: $(uuidgen)" \
  -H "Content-Type: application/json" \
  -d '{"sql":"SELECT * FROM lake.main.\"nyc__taxi_zones\" LIMIT 10"}'

Prefer REST-style filters? We have those too.

# Filter rows PostgREST-style, no SQL needed
curl 'https://api.munimind.com/api/v1/data/nyc.dob_violations/rows?where=bbl.eq.1000010010&limit=100' \
  -H "X-API-Key: $MUNIMIND_API_KEY"

Arrow streaming (Accept: arrow.stream)

Content-negotiate Arrow IPC stream via Accept: application/vnd.apache.arrow.stream for 10-100x smaller payloads. ADBC/pandas/polars/DuckDB read it zero-copy. Same credit cost as JSON.

# Request Apache Arrow IPC stream — 10-100x smaller than JSON,
# zero-copy into pandas / polars / DuckDB.
curl -X POST https://api.munimind.com/api/v1/data/query \
  -H "X-API-Key: $MUNIMIND_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Accept: application/vnd.apache.arrow.stream" \
  -d '{"sql":"SELECT * FROM lake.main.\"nyc__taxi_zones\""}' \
  --output taxi_zones.arrow

# Python — native Arrow reader:
python3 -c "
import pyarrow.ipc as ipc
with open('taxi_zones.arrow','rb') as f:
    reader = ipc.open_stream(f)
    df = reader.read_all().to_pandas()
print(df.head())"

Geospatial — /area/tilequery

# Point-radius GeoJSON — Mapbox-style
curl "https://api.munimind.com/api/v1/area/tilequery?lng=-73.986&lat=40.736&radius_m=250&city=nyc&types=parcels,permits,violations&limit=200" \
  -H "X-API-Key: $MUNIMIND_API_KEY"
# Returns { type: "FeatureCollection", features: [...] } — drop straight
# into Mapbox, Deck.gl, Leaflet, etc. 3 credits / call, 5 km radius cap.

Ask in plain English

Our agent routes between tools, cites every source, and explains its reasoning. Good for investigators doing exploratory work.

curl -X POST https://api.munimind.com/api/v1/answers \
  -H "X-API-Key: $MUNIMIND_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"question":"Who are the top 3 owners on the Lower East Side by HPD violations last year?"}'

Auth + billing

Keys, credits, rate limits

X-API-Key

Every request sends X-API-Key: mm_live_.... Keys scope to a user, tier, and rate limit. Hash-based lookup — we never store keys in plain text. Canary tokens auto-detect compromise.

RateLimit headers

Every response carries IETF-standard RateLimit + RateLimit-Policy + de-facto X-RateLimit-* headers. You know exactly how much budget you have left.