Skip to content

Releases

v0.7.2 — fix hosted Bearer 401 + Accept 406

Choose a tag to compare

@liuyu020923 liuyu020923 released this 09 Jun 11:06

Bug fixes for hosted (remote HTTP) mode

Two bugs caused agents connecting to https://mcp.pangolinfo.com/mcp to fail even with a valid API key:

1. Case-sensitive Bearer → spurious 401

extractApiKey matched Bearer case-sensitively, but many agents/HTTP libraries send lowercase bearer or uppercase BEARER (RFC 7235 says the auth-scheme is case-insensitive). The header was silently ignored, the request fell through to the absent URL param, and the caller got a 401 — surfacing as "my key is correct but I get 401". Now matched with /^bearer\s+(.+)$/i, tolerating extra whitespace.

2. Missing Accept media type → 406

The StreamableHTTP transport requires Accept to advertise both application/json and text/event-stream. Agents frequently send only one, */*, or none. The server now backfills the missing type automatically (patching req.rawHeaders, which the SDK's Hono-based Node adapter actually reads).

Docs

README gains a Hosted (remote HTTP) section documenting the endpoint, the recommended Authorization: Bearer method, the ?api_key= URL fallback, and the Accept-header behavior.


Install (no download needed): point your client at https://mcp.pangolinfo.com/mcp, or use the one-line installer. The attached server.mjs is for manual/local stdio installs.

v0.2.1 — Bug fixes from 3-round SKILL E2E testing

Choose a tag to compare

@liuyu020923 liuyu020923 released this 22 May 17:45

pangolinfo-mcp v0.2.1

Bug fixes uncovered by 3 rounds of independent subagent end-to-end testing of the Lobster GTM SKILL against the cloud MCP. No protocol or API changes — just fixes for incorrect self-reported version + a companion BACKEND-BUGS document for the scrapeapi team.

Fixed

pangolinfo_capabilities was reporting version: "0.1.0" even when running 0.2.0. Hardcoded constant in the tool body wasn't being updated on releases. Now reads from a single src/version.ts source-of-truth, so future bumps update it automatically alongside package.json and README.md.

New

BACKEND-BUGS.md (in the repo, not in the release tarball): 10 issues observed during E2E SKILL testing, prioritized P1→P3, that need scrapeapi backend follow-up. Highlights:

  • B1: filter_categories returns empty for valid categoryIds (data coverage gap)
  • B2: search_categories can't handle multi-token compound keywords
  • B3: wipo_search hol does substring matching — 3-letter brands get massive false-positive hits
  • B4: pointCost field is null in 5+ tools (budget tracking broken for clients)
  • B5: get_amazon_reviews actual cost is 5 credits/page, not 10 as documented

Pangolinfo backend folks: see BACKEND-BUGS.md for full repro + suggested fixes.

Companion SKILL fixes

The "Lobster" GTM SKILL prose (in your Pangolinfo Skills delivery) also has matching corrections — same testing surfaced them:

  • Phase 4 trademark search now uses source="USTM" (US trademark DB) instead of USID (design patent DB). USID has no STATUS field; the prior prose was sending the AI to look for STATUS=Live on records that never had it.
  • Phase 4 new Task 0: resolve consumer brand → legal entity via google_ai_search before querying WIPO. Stanley Quencher's WIPO assignee is "Pacific Market International, LLC", not "STANLEY". Without this step you get unrelated namesake-company hits (Stanley Black & Decker, Victor Stanley, etc.) and miss the actual brand owner.
  • Phase 1 field names corrected: searchVolumeSum → searchVolumeT90, top5BrandsClickShareMax → top5BrandsClickShare, etc. These were aspirational name conventions that don't match the real filter_niches response shape.
  • Reviews cost note: 10 credits/page → ~5 credits/page (matches actual backend behavior; treat response.pointCost as authoritative).

Install

Unchanged from v0.2.0 — stdio downloads from this Release, HTTP runs from the ACR Docker image.

stdio (single-user, local):

mkdir -p ~/.local/lib/pangolinfo-mcp
curl -fsSL https://github.com/pangolinfo/pangolinfo-mcp/releases/download/v0.2.1/server.mjs \
  -o ~/.local/lib/pangolinfo-mcp/server.mjs

HTTP (multi-tenant, cloud):

Aliyun ACR image already pushed: registry-intl.ap-southeast-1.aliyuncs.com/pangolinfo-prod/pangolinfo-mcp:0.2.1. Trigger rolling update in your ACK console (see deploy/DEPLOY-GUIDE.md).

Verify integrity

curl -fsSL https://github.com/pangolinfo/pangolinfo-mcp/releases/download/v0.2.1/server.mjs.sha256 \
  -o server.mjs.sha256
sha256sum -c server.mjs.sha256

SHA256: 693c0514e28494af5e8e94b7530c4f786054739d09ea6b18c4cd8fe57049e05a

Migration from v0.2.0

No code/config changes needed — drop-in replacement. pangolinfo_capabilities will now correctly report "version": "0.2.1". All 17 tools unchanged.

Links

v0.2.0 — HTTP transport for cloud deployment + multi-tenant

Choose a tag to compare

@liuyu020923 liuyu020923 released this 22 May 11:27

pangolinfo-mcp v0.2.0

Big release: adds HTTP transport so a single cloud-deployed instance can serve many users — each request brings its own API key in ?api_key= or Authorization: Bearer. Non-technical users no longer need to install Node + run a local server; they just paste a URL into their AI client.

Why this exists

v0.1.x is stdio-only: every user must install Node 18+, run our installer, have a server.mjs running as a subprocess of their AI client. For technical users (Cursor / Cline / Claude Code) that's fine. For WorkBuddy and other web-based AI agents, and for non-technical users who can't run npm, this is a wall.

v0.2.0 keeps stdio fully working (no breaking change for existing installs) and adds an HTTP / SSE transport that can be deployed to a cloud cluster. Users paste a URL — that's it.

What's new

src/server.ts — dual transport

  • --transport=stdio (default) — same as v0.1.x, AI client forks server.mjs as a subprocess.
  • --transport=http — boots an HTTP server on $PORT (default 3000). Accepts MCP JSON-RPC over POST /mcp, supports SSE streaming. Each request:
    • Extracts API key from ?api_key=pgl_xxx URL query OR Authorization: Bearer pgl_xxx header
    • Builds a per-request PangolinfoClient (no shared auth state between callers)
    • Hands the body to a stateless StreamableHTTPServerTransport instance
  • GET /health and /healthz — k8s-friendly probe endpoint, returns {status, version, toolCount}.
  • Graceful shutdown: SIGTERM/SIGINT triggers a 10s drain window before process.exit(0).

Dockerfile (new)

Multi-stage node:20-alpine → 137 MB runtime image. Non-root user. wget healthcheck. HTTP mode by default.

docker build -t pangolinfo-mcp:0.2.0 .
docker run --rm -p 3000:3000 pangolinfo-mcp:0.2.0

deploy/ (new) — Aliyun ACK manifests

  • k8s-deployment.yaml: 2 replicas, HPA 2→5 @ 70% CPU, liveness/readiness on /health, RollingUpdate maxUnavailable=0.
  • k8s-ingress.yaml: dual-flavor (nginx-ingress / Aliyun ALB) — pick the one matching your cluster's existing setup.
  • DEPLOY-GUIDE.md: end-to-end ACK console walkthrough.

How to use HTTP mode

Cloud (server operator) — deploy the image; expose port 3000 behind a TLS Ingress:

env:
  - name: PANGOLINFO_TRANSPORT
    value: "http"
  - name: PORT
    value: "3000"

Client (your users) — paste the URL into their AI agent:

https://mcp.yourdomain.com/mcp?api_key=pgl_xxxxxxxxxxxx

How to use stdio mode (unchanged)

curl -fsSL https://github.com/pangolinfo/pangolinfo-mcp/releases/download/v0.2.0/server.mjs \
  -o ~/.local/lib/pangolinfo-mcp/server.mjs
# Wire it into your AI client's mcp.json as before

Or via the Pangolinfo Installer — see https://github.com/pangolinfo/pangolinfo-installer

Install

stdio (single-user, local):

mkdir -p ~/.local/lib/pangolinfo-mcp
curl -fsSL https://github.com/pangolinfo/pangolinfo-mcp/releases/download/v0.2.0/server.mjs \
  -o ~/.local/lib/pangolinfo-mcp/server.mjs

HTTP (multi-user, cloud):

docker pull ghcr.io/pangolinfo/pangolinfo-mcp:0.2.0  # (planned)
# or build locally:
git clone https://github.com/pangolinfo/pangolinfo-mcp.git
cd pangolinfo-mcp
docker build -t pangolinfo-mcp:0.2.0 .

Verify integrity

curl -fsSL https://github.com/pangolinfo/pangolinfo-mcp/releases/download/v0.2.0/server.mjs.sha256 \
  -o server.mjs.sha256
sha256sum -c server.mjs.sha256

SHA256: 5f4be4e826fcfc09bd94b2490e9b3d709ff75c7ec2ff5c5869c2b56573c9fcc4

Migration from v0.1.x

No breaking changes. stdio mode is byte-for-byte compatible — same server.mjs filename, same launch command, same behavior. Existing installs upgrade by re-downloading the file.

The HTTP mode is purely additive. If you don't pass --transport=http, nothing changes.

Same 17 tools as v0.1.2

Tool catalog and API are unchanged. See README.md for the full list.

Links

v0.1.2 — surface backend business errors + 9 new Amazon marketplaces

Choose a tag to compare

@liuyu020923 liuyu020923 released this 22 May 09:33

pangolinfo-mcp v0.1.2

Tracks the May 9–22 backend changes in ext-scrapeapi. Two functional improvements:

  1. Business error codes now surface as MCP errors (was: silently passed to the AI as success payload).
  2. 9 new Amazon marketplaces (amz_fr / amz_it / amz_es / amz_ca / amz_au / amz_sa / amz_ae / amz_br / amz_mx) on top of the existing us / uk / de / jp.

Why this release

The Pangolinfo backend returns business errors as HTTP 200 + non-zero code in the JSON envelope (e.g. {code: 4029, message: "Rate limit exceeded", pointCost: 0.0, data: null}). Before v0.1.2, the MCP client only treated non-2xx HTTP as errors — so rate-limit / Akamai-challenge / no-content responses were JSON-stringified into the tool result, and the AI would misinterpret them as "scrape succeeded but returned no data." With the recent backend changes (May 9–22) adding three more such codes (4029 rate-limit refunded, 9200 Akamai detected, 9201 no-other-sellers), this gap became urgent.

Changes

src/client.ts — after a 2xx HTTP response, inspect the JSON envelope:

Backend code Meaning MCP error class
0 success (no error, return payload)
4001 / 4002 / 4003 auth / IP denied / permission AUTH
4029 rate-limit (refunded) RATE_LIMIT
9200 scrape no-content (incl. Akamai challenge, refunded) SERVER (retryable)
9201 ASIN has no other sellers BAD_INPUT
Other 4xxx invalid request BAD_INPUT
5xxx / 9xxx server / scrape failure SERVER

Errors surface to the AI as isError: true with a translated hint — the AI now knows to retry / change input / give up, rather than treating an error envelope as empty data.

Site enums (9 tools) — search_amazon, get_amazon_product, get_amazon_reviews, list_bestsellers, list_new_releases, list_seller_products, list_category_products, search_categories, get_category_paths now accept all 13 marketplaces the backend supports:

amz_us  amz_uk  amz_de  amz_jp
amz_fr  amz_it  amz_es  amz_ca  amz_au
amz_sa  amz_ae  amz_br  amz_mx

SITE_TO_DOMAIN maps for the 3 URL-building tools (search_amazon, get_amazon_product, get_amazon_reviews) are aligned with the enum, so amz_fr → www.amazon.fr etc. resolve correctly.

zipcode description — updated across 7 tools to spell out the new backend rule:

ZIP code that must match the site country (amz_us → US zip, amz_jp → JP zip, ...). Optional; backend picks a random one from the per-country pool when omitted. Cross-country zips (e.g. amz_us + JP zip) are rejected by the backend.

This tracks the May 22 backend tightening (commit 0e3484b2): zipcode is now strictly validated against the host's country whitelist.

Install

Same as v0.1.1 — single self-contained server.mjs, no npm install.

macOS / Linux

mkdir -p ~/.local/lib/pangolinfo-mcp
curl -fsSL https://github.com/pangolinfo/pangolinfo-mcp/releases/download/v0.1.2/server.mjs \
  -o ~/.local/lib/pangolinfo-mcp/server.mjs

Windows (PowerShell)

$dir = "$env:LOCALAPPDATA\pangolinfo-mcp"
New-Item -ItemType Directory -Force -Path $dir | Out-Null
irm https://github.com/pangolinfo/pangolinfo-mcp/releases/download/v0.1.2/server.mjs `
  -OutFile "$dir\server.mjs"

Verify integrity

curl -fsSL https://github.com/pangolinfo/pangolinfo-mcp/releases/download/v0.1.2/server.mjs.sha256 \
  -o server.mjs.sha256
sha256sum -c server.mjs.sha256
# expect: server.mjs: OK

SHA256: 9dcfc0ee16aaf11d7ce98da54a8ccd2f510b9009e5f086752a10f1ac3ab83e89

Migration from v0.1.1

  • No config changes — same server.mjs filename, same launch command.
  • AI behavior changes: tools that previously returned {code: 4029, ...} payloads now throw a proper RATE_LIMIT error. If you had AI prompts that pattern-matched on data: null, switch to handling the isError: true envelope.
  • New amz_* site values are additive — existing amz_us / amz_uk / amz_de / amz_jp calls are unchanged.

Links

v0.1.1 — eliminate Node ESM warning (rename to server.mjs)

Choose a tag to compare

@liuyu020923 liuyu020923 released this 22 May 09:15

pangolinfo-mcp v0.1.1

Patch release: rename release artifact server.js → server.mjs so Node.js recognizes it as an ES module without a sidecar package.json. Functionally identical to v0.1.0.

Why this patch

When users downloaded server.js from v0.1.0 and ran it standalone, Node printed:

Warning: Module type of file://.../server.js is not specified and it doesn't parse
as CommonJS. Reparsing as ES module because module syntax was detected. This incurs
a performance overhead.

The warning goes to stderr — harmless, but noisy in AI client logs and adds 1-2 ms parse latency on every start. Renaming to .mjs makes Node treat the file as ESM unambiguously, eliminating the warning entirely.

Install

Single self-contained file (~800 KB), no npm install.

macOS / Linux

mkdir -p ~/.local/lib/pangolinfo-mcp
curl -fsSL https://github.com/pangolinfo/pangolinfo-mcp/releases/download/v0.1.1/server.mjs \
  -o ~/.local/lib/pangolinfo-mcp/server.mjs

Windows (PowerShell)

$dir = "$env:LOCALAPPDATA\pangolinfo-mcp"
New-Item -ItemType Directory -Force -Path $dir | Out-Null
irm https://github.com/pangolinfo/pangolinfo-mcp/releases/download/v0.1.1/server.mjs `
  -OutFile "$dir\server.mjs"

Or use the Pangolinfo Installer one-liner to auto-configure 7 supported AI clients.

Verify integrity

curl -fsSL https://github.com/pangolinfo/pangolinfo-mcp/releases/download/v0.1.1/server.mjs.sha256 \
  -o server.mjs.sha256
sha256sum -c server.mjs.sha256
# expect: server.mjs: OK

SHA256: 8ffc37a17763b786cca55034a5e881066a56ed86daf63b941abf2bf7cf5dd14a

(Same byte content as v0.1.0 server.js — the file was renamed, not rebuilt with code changes.)

Migration from v0.1.0

If you wired v0.1.0 manually, update the args path in your AI client's MCP config:

- "args": ["/abs/path/to/server.js"]
+ "args": ["/abs/path/to/server.mjs"]

Installer users get this for free on the next install run.

What's in this release

Same 17 tools as v0.1.0 — see README.md for the full table.

Supported AI clients: Claude Code, Cursor, Cline (VS Code), Windsurf, Codex, Hermes, OpenClaw.

Requires: Node.js 18+. Each Amazon scrape call costs 0.75 credits; pangolinfo_capabilities is free.

Get an API key

Sign up at https://extapi.pangolinfo.com.

Links

v0.1.0 — 17 Amazon e-commerce tools for AI assistants

Choose a tag to compare

@liuyu020923 liuyu020923 released this 22 May 09:12

pangolinfo-mcp v0.1.0

First public release of the Pangolinfo MCP server — 17 Amazon e-commerce data tools for AI assistants via Model Context Protocol.

Install

The release ships a single self-contained server.js (~800 KB) — all dependencies are bundled in. No npm install needed.

macOS / Linux

mkdir -p ~/.local/lib/pangolinfo-mcp
curl -fsSL https://github.com/pangolinfo/pangolinfo-mcp/releases/download/v0.1.0/server.js \
  -o ~/.local/lib/pangolinfo-mcp/server.js

Windows (PowerShell)

$dir = "$env:LOCALAPPDATA\pangolinfo-mcp"
New-Item -ItemType Directory -Force -Path $dir | Out-Null
irm https://github.com/pangolinfo/pangolinfo-mcp/releases/download/v0.1.0/server.js `
  -OutFile "$dir\server.js"

Or use the Pangolinfo Installer one-liner to auto-configure 7 supported AI clients.

Verify download integrity

curl -fsSL https://github.com/pangolinfo/pangolinfo-mcp/releases/download/v0.1.0/server.js.sha256 \
  -o server.js.sha256
sha256sum -c server.js.sha256
# expect: server.js: OK

SHA256: 8ffc37a17763b786cca55034a5e881066a56ed86daf63b941abf2bf7cf5dd14a

What's in this release

17 tools (see README.md for the full table):

  • Amazon: search_amazon, get_amazon_product, get_amazon_reviews, list_bestsellers, list_new_releases, list_seller_products, list_category_products
  • Categories: search_categories, get_category_children, filter_categories, filter_niches, get_category_paths
  • Other data sources: search_local_maps, wipo_search, google_ai_search, google_trends
  • Self-introspection: pangolinfo_capabilities (free, 0 credits)

Supported AI clients: Claude Code, Cursor, Cline (VS Code), Windsurf, Codex, Hermes, OpenClaw — per-client config snippets are in the README.

Languages: Chinese (default) and English tool descriptions + error hints. Set PANGOLINFO_LANG=en to switch.

Requires: Node.js 18+. Each Amazon scrape call costs 0.75 credits; pangolinfo_capabilities is free.

Get an API key

Sign up at https://extapi.pangolinfo.com.

Links