io.github.chrischall/redfin-mcp

Fetch up to 200 Redfin property records in a single tool call. Provide an array of targets, each one of: a `url` (full Redfin homedetails URL or path with the /home/<id> segment), a `property_id` alone (resolved internally by following Redfin's /home/<id> redirect to the canonical listing), or a `property_id`+`listing_id` pair (fastest — skips resolution). Returns the same per-property record shape as `redfin_get_property`, but without a summary table — use `redfin_compare_properties` for that. Per-target errors are captured per-row; a single bad ID does not fail the batch. Server-side concurrency, ~6 in flight at a time, with retry-once-on-timeout per row to absorb transient bridge hiccups. The whole call is bounded by an overall hard deadline: a single slow/hung row never wedges the server — when the deadline is reached any unsettled row is returned with `status: "pending"` (retryable) and a `pending` count so you can re-run just those targets. Use this when you have a list of saved homes / candidate properties and need the full structured data for every one of them.

Solve for the maximum home price you can afford under the standard 28/36 DTI rule. Inputs: monthly income, recurring monthly debts (car loans, student loans, etc.), down payment, interest rate, and optional property-tax rate / insurance / HOA / loan term. Output: max home price, the binding constraint (front-end vs back-end), and the full PITI breakdown at that price. Same math across the realty MCP cohort (@chrischall/realty-core). No network — pure local math.

Local-only mortgage payment calculator. Returns a full PITI breakdown (principal + interest, property tax, insurance, HOA, PMI) and total interest over the life of the loan. No network call — fully deterministic, safe to use for scenario comparison without burning a fetch. Provide either `down_payment` OR `down_payment_percent`; defaults to 20%. Property tax can be given as `property_tax_annual` or `property_tax_rate` (% of home price). PMI applies automatically when LTV > 80% and `pmi_rate` is provided.

Fetch and compare 2 to 25 Redfin properties side-by-side. Provide an array of targets, each either a `url` or a `property_id`+`listing_id` pair. Returns the full per-property record (price, beds/baths, sqft, year built, HOA monthly, last sold, derived price-drop, etc.). For >25 properties or workflows that don't need side-by-side analysis use `redfin_bulk_get`. Pass `include_summary: true` for an aligned-by-field `summary` table (default false to save context — the per-row records carry the same data, so emitting both duplicates ~30% of the response weight). Each record's `extracted_features` (lake_front, hot_tub, basement, furnished, dock, community) is always included. The raw marketing description is omitted by default — opt in with `include_description: true`. Errors for individual properties are captured per-row with a `status` (`ok` / `timeout` / `bridge_down` / `protocol` / `pending` / `other`) and `retryable`. Server-side concurrency (~6 in flight) with retry-once-on-timeout per row; the whole call is bounded by an overall deadline, and any row still unsettled then comes back as `status: "pending"` (retryable) with a `pending` count.

Fetch climate risk for a small set of representative URLs in an area, then return their averaged baseline values plus the shared cluster_id when present. Use this as a cheap area-level read BEFORE fanning out a per-property call: if all sample properties agree (same cluster_id, same fire/flood/heat factors), the baseline applies to the whole cluster and N redundant fetches are avoidable. Pass 2–10 URLs you believe represent the area; returns the aggregate plus the per-URL responses for transparency. Server-side concurrency (~5 fetches in flight) with retry-once-on-timeout per URL; the whole call is bounded by an overall deadline, and any sample still unsettled then comes back as `status: "pending"` (retryable) with a `pending` count and is left out of the averages. Limitations of the per-property tool apply (no landslide coverage — note documented in the per-property tool description).

Resolve a free-text address (with optional city/state/zip) to its Redfin canonical home URL and home_id. Walks a 3-rung ladder: (1) autocomplete as-typed, (2) autocomplete with suffix expansion (Rd ↔ Road, Ln ↔ Lane, etc.), (3) search fallback (#75) — when autocomplete misses entirely and city/state are provided, resolves the locality to a region, fires a bounded gis search, and fuzzy-matches the input street tokens against returned homes. `matched_via` is `'autocomplete'` or `'search_fallback'`. A miss on every rung returns `resolved: false` (with the `attempts` tried) rather than an error; a signed-out session, bridge failure or region mismatch is still reported as an error. Address discrepancies across MLS feeds are common (the `109 vs 169 Overlook Point Ln` cross-MLS case is a regular occurrence) — `redfin_get_property` on the resolved URL surfaces them as `address_alternates[]` (#42). Use this when you have a property address and need its Redfin home_id for follow-on calls (e.g. `redfin_get_property`). Read-only, no auth required.

First Street Foundation climate risk scores for a property. COVERS: flood (factor 1–10, FEMA zones, 30-year annual chance series), fire (factor 1–10, relative risk, insurance price band, provider count), heat (factor 1–10, cumulative-risk projections at 0/5/10/15/20-year horizons). DOES NOT COVER: landslide. This is the Helene-relevant risk vector in the NC mountains market and many parts of California / the Pacific Northwest. First Street has no landslide product — for that vector check the NC Geological Survey landslide hazard maps (NC-specific) or USGS landslide hazard data (national). Surfaced on every response as `not_covered: ['landslide']`. Response shape: when First Street data is available, `available: true` with the risk blocks; when not, `{ available: false, reason }` where reason is one of `no_first_street_data`, `new_construction`, `address_outside_coverage`. Sourced from Redfin's server-rendered homedetails HTML (no clean stingray endpoint exists). Pass a homedetails URL (full or path) — it must end in a `/home/<propertyId>` segment; any other page is rejected without being fetched. When a `cluster_id` is surfaced, properties with the same value typically share identical climate scores — use that to group N properties and skip redundant fetches.

Fetch climate risk for up to 100 property URLs in a single call. Same per-property shape as `redfin_get_climate_risk`; output preserves input order. Per-row error capture — properties without First Street data return `{ available: false, reason }` without aborting the batch. Server-side concurrency (~5 fetches in flight) with retry-once-on-timeout per URL. The whole call is bounded by an overall deadline: any URL still unsettled then comes back as `status: "pending"` (retryable) with a `pending` count, so re-run just those. Use this when batching ~60-property workflows where climate risk is the dominant cost. Limitations from the per-property tool apply (no landslide coverage).

Find nearby rental comparables for a given property: nearby active rental listings with similar bed/bath/sqft, including monthly rent, distance, and the Redfin URL. Useful for estimating what a property could rent for, or for finding rentals near a home you're considering. Inputs are the rent estimate range + lat/lng + propertyId. `redfin_get_property` supplies property_id, latitude and longitude, but it does not return a rent estimate — take the range from the Redfin property page's rental estimate or another source (e.g. a Zillow rent Zestimate), or pass a deliberately wide range around a known rent. The range is sent to Redfin as a filter, so a wrong or invented range narrows the comps; low must not exceed high.

Market report for a Redfin region: median list/sold prices, $/sqft, sale-to-list ratio, total homes for sale + sold, all with year-over-year and month-over-month change. Provide either (a) `location` — free-text we resolve via autocomplete (best with city names; "New York", "Seattle"; neighborhoods typically return empty data), or (b) `region_id` + `region_type` directly. `property_type` defaults to 1 (all). Each metric returns `{ label, value, unit, yoy_change_fraction, yoy_direction, mom_change_fraction, mom_direction }`. Read-only; safe to call repeatedly.

Listing-price events for a property — listings, price changes, pending, sold, etc. Each entry has a date, event description, price, days-on-market at that point, and the data-source attribution (MLS, county records, etc.). Also returns the tax-history series from public records. Provide either `url` or `property_id`+`listing_id`. Read-only; safe to call repeatedly.

Fetch a property's full Redfin record. Provide one of: (a) `url` — full Redfin homedetails URL or path, resolved via the initialInfo endpoint; (b) `property_id` alone — resolved internally by following Redfin's /home/<id> redirect to the canonical listing, then initialInfo; or (c) `property_id` + `listing_id` — fastest, skips resolution and goes straight to aboveTheFold. Returns address, beds/baths, sqft, lot_size (sq ft), year built, price, status, days on market, plus derived fields (lot_size_acres, price_drop_*, hoa_monthly_usd, last_sold_*, tax_annual, extracted_features). lot_size / lot_size_acres are null (never 0) for condos and listings with no public-records lot. `primary_photo_url` is a raw Redfin CDN URL and is dropped on the default `compact` view — pass `view: "full"` for it, or use `redfin_get_property_photos` for the whole gallery. The raw marketing description is OMITTED by default — opt in with `include_description: true`. Set `include_price_history: true` to bundle the full price history (and the cross-MCP-normalized `events_normalized` view) inline; set `include_tax_history: true` for `tax_history`. Read-only; safe to call repeatedly.

The full photo gallery for a Redfin property — every image in mediaBrowserInfo. Each entry returns CDN URLs at multiple sizes (fullscreen, large, medium, lightbox) plus a thumbnail and the photo's caption when set. Provide either `url` (full Redfin homedetails URL or path; we resolve to IDs via initialInfo) or `property_id` + `listing_id` (skip the resolve step). Returns `{ property_id, listing_id, count, photos }`. Off-market or stub listings may return count=0. Read-only; safe to call repeatedly.

The signed-in user's favorited homes on redfin.com. Returns `{ count, homes }` — each home has address, price, beds/baths, status — plus a `note` when none were found (no favorites vs. a page-scrape miss). Requires the user to be signed in. Read-only; safe to call repeatedly.

The signed-in user's saved searches on redfin.com, derived from the saved-searches page HTML. Returns `{ count, searches }`; each entry is `{ url, region_segment, display_text }`. Requires the user to be signed in. When none are found, `searches` is empty and a `note` explains it may be no saved searches or a page-scrape miss. Read-only; safe to call repeatedly.

List every registered account label plus the current `active_session_id`. When none are registered, `sessions` is empty and `active_session_id` is null. The registry is a label only: it does not change which Redfin account requests use. Every redfin tool call goes through whichever browser tab the ContextMint Bridge extension is signed into; to read a different account, sign that tab into it.

Round-trips a small public www.redfin.com URL (/robots.txt) through ContextMint Bridge (your signed-in browser tab) and returns diagnostics: the bridge's role (host/peer/null), port, version, the extension link (linked / pair pending / not attached / never answered), the elapsed round-trip time, and a plain-English hint distinguishing 'bridge never came up' from 'extension not connected' from 'this browser can't serve a capability' from 'real www.redfin.com-side problem'. Read-only, no auth required. Call this when a real tool fails and you want to know which hop broke.

Record (or refresh) which Redfin account the signed-in session belongs to, keyed by `account_identity`. Re-registering the same identity updates the existing entry. Returns a `session_id`; pass `mark_active: true` to mark it as the current account in the same call. The registry is a label only: it does not change which Redfin account requests use. Every redfin tool call goes through whichever browser tab the ContextMint Bridge extension is signed into; to read a different account, sign that tab into it.

Resolve up to 100 free-text street addresses to Redfin canonical home URLs + home_ids in a single tool call. Each input is either a string (full address) or a structured `{street, city, state, zip}` object. Output preserves input order. Unresolved entries return `resolved: false` without aborting the batch; a transient bridge failure surfaces a distinct retryable `status` (timeout/bridge_down/pending) so it is never mistaken for a genuine no-match. Per-row retry-once-on-timeout, server-side concurrency ~6 in flight. The whole call is bounded by an overall hard deadline: a single slow/hung row never wedges the server — unsettled rows come back with `status: "pending"` and a `pending` count so you can re-run just those. Use this when you have a list of properties from another system (Compass, MLS, spreadsheet) and need their Redfin handles for follow-on calls — collapses the typical 6-search-call + 15-resolve flow into one trip.

Search Redfin listings by location (city, ZIP, neighborhood, or full street address) and optional filters. Resolves the location via Redfin's autocomplete then queries the gis API; full street addresses short-circuit to the single matched home (no gis call). Returns matching properties with price, beds/baths, sqft, year built, address, and the Redfin home URL. `resolved_as` is `'region'` / `'address'`. `coverage` is `'full'` (gis indexed this region), `'profile_only'` (Redfin has profiles for individual addresses here but search isn't indexed — use redfin_get_by_address per property), or `'none'`. `result_cap_hit: true` signals the result page is full — gis returned as many rows as requested (`limit`, default 40, max 350) — so more listings likely exist; raise `limit` or narrow with price/beds filters. ZIP queries that fall into Redfin's cross-continent fallback (e.g. ZIP 28746 returning Seattle results) now error loudly. Only `for_sale` status is supported; `sold` / `for_rent` return an error (use redfin_get_comparable_rentals for rentals, redfin_get_market_report for sold-price trends). Read-only; safe to call repeatedly.

Mark a previously registered `session_id` as the current account label, for your own bookkeeping across a conversation. The registry is a label only: it does not change which Redfin account requests use. Every redfin tool call goes through whichever browser tab the ContextMint Bridge extension is signed into; to read a different account, sign that tab into it.

Enumerate EVERY for-sale Redfin listing inside a bounding box without silent truncation. Redfin's gis API returns at most 350 homes per call and ignores server-side filters, so this tool searches drawn-map polygons, recursively quarters any tile that hits the cap, dedupes by property_id and re-applies the caller's filters locally. Homes outside the box and tiles where Redfin drifted to another region are left out. Alternatively pass `zips` (ZIP mode) to sweep a list of ZIP regions instead of map tiles, when Redfin ignores drawn-map polygons. Returns a completeness summary (requests, tiles or ZIPs, unique listings, tiles/ZIPs still at the cap, drift warnings) plus the listings inline by default; pass the optional `output_path` to write them to a new JSON file instead, for large areas (inside the output directory: REDFIN_OUTPUT_DIR, default ~/Downloads/redfin-mcp). Sequential requests with a delay. Read-only against Redfin; the only write is the optional local output file.