Get started/Changelog

    Get started

    API Changelog

    Every change to what the scanr.ai API and MCP server accept or return, newest first. Changes to the scanr.ai app itself are in the app changelog.

    Each entry is dated, and that date is the contract version. The OpenAPI spec publishes it as info.version, so you can tell which entry the API you're calling matches. /api/v2 changes only for a redesign, never for an entry here.

    Each change carries one tag:

    TagMeaning
    BreakingA client that worked before may now fail or behave differently. We email every API subscriber before it ships.
    DeprecatedStill works, but has a replacement and will be removed. Responses carry a Deprecation header, and a Sunset header once there's a removal date.
    ChangedBehaves differently in a way that shouldn't break a client: a fix, a new limit you're unlikely to reach, a clearer error.
    AddedNew endpoints, fields, parameters or MCP tools.

    Deprecation policy

    A deprecated endpoint or field keeps working for at least 90 days after its entry here. Its responses carry Deprecation and Link: <…/changelog>; rel="deprecation" headers from the day it's announced, and a Sunset header with the removal date once one is set. Its removal gets a Breaking entry of its own.

    2026-11-15

    • Changed: The technical indicator endpoints (POST /api/v2/technicals/technical/bulk, the per-indicator routes and relative-volume-stdev) now apply the breadth symbol rules to a $ ticker such as $S2-L1000, the same rules as its bars. The timeframe must be daily or longer (400 otherwise). API access needs an Ultimate token, and in the app the stage, unusual-volume and compression series need Super Pro (403 otherwise).
    • Changed: An indicator that can't be computed over a breadth symbol is now a 422 with the scanner's code, where it used to be a 500. That covers anything reading volume (scan.supplied_series.no_volume), a feature window or group series (scan.supplied_series.input_unavailable), and an intraday timeframe (stocks.breadth_symbol_end_of_day).

    2026-11-11

    • Changed: The Business plan is now called Ultimate. Its limits, datasets and price are unchanged, and existing tokens keep working. Error messages, the OpenAPI spec and these docs use the new name.
    • Breaking: cache_ttl_seconds on POST /api/v2/scan and /scan/stream now depends on your API plan. Starter always uses 120 seconds, Power User can set 30 to 900, and Ultimate can set 0 to 900. A value outside your plan's range is clamped to it, not rejected, so a Starter request that sent 0 (no cache) now gets results up to four minutes old: two minutes fresh, then two more while a fresh one is computed. See Caching.
    • Changed: A short cache_ttl_seconds is now honored whoever ran the same payload before you. Before, a result cached by a request with a longer window could be returned to you as fresh, up to twice that window old.
    • Added: max_stale_seconds on POST /api/v2/scan and /scan/stream (Power User and Ultimate). It sets how much older than cache_ttl_seconds a result can be while a fresh one is computed. 0 never returns a result older than cache_ttl_seconds.
    • Added: A cache block on scan responses: the ttl_seconds and max_stale_seconds that applied, the result's age_seconds, and whether it was stale.
    • Added: cache_ttl_seconds and max_stale_seconds arguments on the scan MCP tool, with the same plan limits. Before, every MCP scan used the 120-second default.

    2026-11-08

    • Breaking: POST /api/v2/scan and the scan MCP tool now share one rate limit. On Starter that's 30 scans a minute across both, where each used to allow 30. Other endpoints and MCP tools keep their own counters. See Plans & Rate Limits.
    • Breaking: Daily AI credits for natural-language scans (POST /api/v2/scan/natural, /scan/refine, and the natural_scan and refine_payload MCP tools) are now 25,000 on Starter, 50,000 on Power User and 100,000 on Business, down from 45,000, 180,000 and 900,000. That's about 300, 600 and 1,200 natural-language scans a day. Running a scan payload still uses no AI credits. See AI credits.
    • Deprecated: POST /api/v2/scan/kraken-scan-results-attachments. Use POST /api/v2/scan/saved-scan-results, which takes the same body and returns the same response. The old path keeps working and now answers with a Deprecation header. It has no removal date yet.
    • Changed: A scan payload that fails validation no longer counts against your scan rate limit, on POST /api/v2/scan, /scan/stream or the scan MCP tool.
    • Changed: POST /api/v2/scan/stream now applies your API plan the same way POST /api/v2/scan does. Before, a token could be refused or have results limited on the stream but not on /scan.
    • Changed: POST /api/v2/scan/natural and /scan/refine wait up to 120 seconds for an answer, and are no longer retried behind the scenes.
    • Changed: GET /api/v2/scan/metrics-popularity no longer returns 500 to API tokens, and MCP calls to mcp.scanr.ai no longer return 404.
    • Added: OAuth 2.1 sign-in for the API and the MCP server, so an MCP client or app can connect without a pasted token. See Authentication.
    • Added: POST /api/v2/scan/saved-scan-results, the documented path for saved-scan results.
    • Added: The stage_history MCP tool, returning a stock's weekly Weinstein stage, sub-stage, weeks in stage and Mansfield relative strength. See the stage analysis guide.