x-twitter-scraper

name	x-twitter-scraper
description	X API & Twitter scraper skill for AI coding agents. Builds integrations with the Xquik REST API, MCP server & webhooks: tweet search, user lookup, follower extraction, engagement metrics, giveaway contest draws, trending topics, account monitoring, reply/retweet/quote extraction, community & Space data, mutual follow checks, write actions (tweet, like, retweet, follow, DM, profile, media upload, communities), Telegram integrations. Works with Claude Code, Cursor, Codex, Copilot, Windsurf & 40+ agents.
compatibility	Requires internet access to call the Xquik REST API (https://xquik.com/api/v1)
license	MIT
metadata	{"author":"Xquik","version":"1.7.0"}

Xquik API Integration

Xquik is an X (Twitter) real-time data platform providing a REST API, HMAC webhooks, and an MCP server for AI agents. It covers account monitoring, bulk data extraction (20 tools), giveaway draws, tweet/user lookups, media downloads, follow checks, trending topics, write actions (tweet, like, retweet, follow, DM, profile, media upload, communities), and Telegram integrations.

Quick Reference

Base URL	https://xquik.com/api/v1
Auth	x-api-key: xq_... header (64 hex chars after xq_ prefix)
MCP endpoint	https://xquik.com/mcp (StreamableHTTP, same API key)
Rate limits	10 req/s sustained, 20 burst (API); 60 req/s sustained, 100 burst (general)
Pricing	$20/month base (1 monitor included), $5/month per extra monitor
Quota	Monthly usage cap. 402 when exhausted. Enable extra usage from dashboard for overage (tiered spending limits: $5/$7/$10/$15/$25)
Docs	docs.xquik.com
HTTPS only	Plain HTTP gets 301 redirect

Authentication

Every request requires an API key via the x-api-key header. Keys start with xq_ and are generated from the Xquik dashboard. The key is shown only once at creation; store it securely.

const API_KEY = "xq_YOUR_KEY_HERE";
const BASE = "https://xquik.com/api/v1";
const headers = { "x-api-key": API_KEY, "Content-Type": "application/json" };

For Python examples, see references/python-examples.md.

Choosing the Right Endpoint

Goal	Endpoint	Notes
Get a single tweet by ID/URL	GET /x/tweets/{id}	Full metrics: likes, retweets, views, bookmarks, author info
Search tweets by keyword/hashtag	GET /x/tweets/search?q=...	Tweet info with optional engagement metrics (likeCount, retweetCount, replyCount)
Get a user profile	GET /x/users/{username}	Name, bio, follower/following counts, profile picture, location, created date, statuses count
Check follow relationship	GET /x/followers/check?source=A&target=B	Both directions
Get trending topics	GET /trends?woeid=1	Regional trends by WOEID. Metered
Get radar (trending news)	GET /radar?source=hacker_news	Free, 7 sources: Google Trends, Hacker News, Polymarket, TrustMRR, Wikipedia, GitHub, Reddit
Monitor an X account	POST /monitors	Track tweets, replies, quotes, retweets, follower changes
Update monitor event types	PATCH /monitors/{id}	Change subscribed events or pause/resume
Poll for events	GET /events	Cursor-paginated, filter by monitorId/eventType
Receive events in real time	POST /webhooks	HMAC-signed delivery to your HTTPS endpoint
Update webhook	PATCH /webhooks/{id}	Change URL, event types, or pause/resume
Run a giveaway draw	POST /draws	Pick random winners from tweet replies
Download tweet media	POST /x/media/download	Single (tweetInput) or bulk (tweetIds[], up to 50). Returns gallery URL. First download metered, cached free
Extract bulk data	POST /extractions	20 tool types, always estimate cost first
Check account/usage	GET /account	Plan status, monitors, usage percent
Link your X identity	PUT /account/x-identity	Required for own-account detection in style analysis
Analyze tweet style	POST /styles	Cache recent tweets for style reference
Save custom style	PUT /styles/{username}	Save custom style from tweet texts (free)
Get cached style	GET /styles/{username}	Retrieve previously cached tweet style
Compare styles	GET /styles/compare?username1=A&username2=B	Side-by-side comparison of two cached styles
Get tweet performance	GET /styles/{username}/performance	Live engagement metrics for cached tweets
Save a tweet draft	POST /drafts	Store drafts for later
List/manage drafts	GET /drafts, DELETE /drafts/{id}	Retrieve and delete saved drafts
Compose a tweet	POST /compose	3-step workflow (compose, refine, score). Free, algorithm-backed
Connect an X account	POST /x/accounts	Credentials encrypted at rest. Required for write actions
List connected accounts	GET /x/accounts	Free
Re-authenticate account	POST /x/accounts/{id}/reauth	When session expires
Post a tweet	POST /x/tweets	From a connected account. Supports replies, media, note tweets, communities
Delete a tweet	DELETE /x/tweets/{id}	Must own the tweet via connected account
Like / Unlike a tweet	POST / DELETE /x/tweets/{id}/like	Metered
Retweet	POST /x/tweets/{id}/retweet	Metered
Follow / Unfollow a user	POST / DELETE /x/users/{id}/follow	Metered
Send a DM	POST /x/dm/{userId}	Text, media, reply to message
Update profile	PATCH /x/profile	Name, bio, location, URL
Upload media	POST /x/media	FormData. Returns media ID for tweet attachment
Community actions	POST /x/communities, POST /x/communities/{id}/join	Create, delete, join, leave
Create Telegram integration	POST /integrations	Receive monitor events in Telegram. Free
Manage integrations	GET /integrations, PATCH /integrations/{id}	List, update, delete, test, deliveries. Free

See references/mcp-tools.md for tool selection rules, common mistakes, and unsupported operations.

Error Handling & Retry

All errors return { "error": "error_code" }. Key error codes:

Status	Code	Action
400	invalid_input, invalid_id, invalid_params, invalid_tweet_url, invalid_tweet_id, invalid_username, invalid_tool_type, invalid_format, missing_query, missing_params, webhook_inactive, no_media	Fix the request, do not retry
401	unauthenticated	Check API key
402	no_subscription, subscription_inactive, usage_limit_reached, no_addon, extra_usage_disabled, extra_usage_requires_v2, frozen, overage_limit_reached	Subscribe, enable extra usage, or wait for quota reset
403	monitor_limit_reached, api_key_limit_reached	Delete a monitor/key or add capacity
404	not_found, user_not_found, tweet_not_found, style_not_found, draft_not_found, account_not_found	Resource doesn't exist or belongs to another account
409	monitor_already_exists, account_already_connected, already_member	Resource already exists, use the existing one
422	connection_failed, reauth_failed	X credential verification failed. Check credentials
429	x_api_rate_limited	Rate limited. Retry with exponential backoff, respect Retry-After header
500	internal_error	Retry with backoff
502	stream_registration_failed, x_api_unavailable, x_api_unauthorized, x_write_failed, upstream_error, delivery_failed	Retry with backoff

Retry only 429 and 5xx. Never retry 4xx (except 429). Max 3 retries with exponential backoff:

async function xquikFetch(path, options = {}) {
  const baseDelay = 1000;

  for (let attempt = 0; attempt <= 3; attempt++) {
    const response = await fetch(`${BASE}${path}`, {
      ...options,
      headers: { ...headers, ...options.headers },
    });

    if (response.ok) return response.json();

    const retryable = response.status === 429 || response.status >= 500;
    if (!retryable || attempt === 3) {
      const error = await response.json();
      throw new Error(`Xquik API ${response.status}: ${error.error}`);
    }

    const retryAfter = response.headers.get("Retry-After");
    const delay = retryAfter
      ? parseInt(retryAfter, 10) * 1000
      : baseDelay * Math.pow(2, attempt) + Math.random() * 1000;

    await new Promise((resolve) => setTimeout(resolve, delay));
  }
}

Cursor Pagination

Events, draws, extractions, and extraction results use cursor-based pagination. When more results exist, the response includes hasMore: true and a nextCursor string. Pass nextCursor as the after query parameter.

async function fetchAllPages(path, dataKey) {
  const results = [];
  let cursor;

  while (true) {
    const params = new URLSearchParams({ limit: "100" });
    if (cursor) params.set("after", cursor);

    const data = await xquikFetch(`${path}?${params}`);
    results.push(...data[dataKey]);

    if (!data.hasMore) break;
    cursor = data.nextCursor;
  }

  return results;
}

Cursors are opaque strings. Never decode or construct them manually.

Extraction Tools (20 Types)

Extractions run bulk data collection jobs. The complete workflow: estimate cost, create job, retrieve results, optionally export.

Tool Types and Required Parameters

Tool Type	Required Field	Description
reply_extractor	targetTweetId	Users who replied to a tweet
repost_extractor	targetTweetId	Users who retweeted a tweet
quote_extractor	targetTweetId	Users who quote-tweeted a tweet
thread_extractor	targetTweetId	All tweets in a thread
article_extractor	targetTweetId	Article content linked in a tweet
follower_explorer	targetUsername	Followers of an account
following_explorer	targetUsername	Accounts followed by a user
verified_follower_explorer	targetUsername	Verified followers of an account
mention_extractor	targetUsername	Tweets mentioning an account
post_extractor	targetUsername	Posts from an account
community_extractor	targetCommunityId	Members of a community
community_moderator_explorer	targetCommunityId	Moderators of a community
community_post_extractor	targetCommunityId	Posts from a community
community_search	targetCommunityId + searchQuery	Search posts within a community
list_member_extractor	targetListId	Members of a list
list_post_extractor	targetListId	Posts from a list
list_follower_explorer	targetListId	Followers of a list
space_explorer	targetSpaceId	Participants of a Space
people_search	searchQuery	Search for users by keyword
tweet_search_extractor	searchQuery	Search and extract tweets by keyword or hashtag (bulk, up to 1,000)

Complete Extraction Workflow

// Step 1: Estimate cost before running (pass resultsLimit if you only need a sample)
const estimate = await xquikFetch("/extractions/estimate", {
  method: "POST",
  body: JSON.stringify({
    toolType: "follower_explorer",
    targetUsername: "elonmusk",
    resultsLimit: 1000, // optional: limit to 1,000 results instead of all
  }),
});
// Response: { allowed: true, estimatedResults: 195000000, usagePercent: 12, projectedPercent: 98 }

if (!estimate.allowed) {
  console.log("Extraction would exceed monthly quota");
  return;
}

// Step 2: Create extraction job (pass same resultsLimit to match estimate)
const job = await xquikFetch("/extractions", {
  method: "POST",
  body: JSON.stringify({
    toolType: "follower_explorer",
    targetUsername: "elonmusk",
    resultsLimit: 1000,
  }),
});
// Response: { id: "77777", toolType: "follower_explorer", status: "completed", totalResults: 195000 }

// Step 3: Poll until complete (large jobs may return status "running")
while (job.status === "pending" || job.status === "running") {
  await new Promise((r) => setTimeout(r, 2000));
  job = await xquikFetch(`/extractions/${job.id}`);
}

// Step 4: Retrieve paginated results (up to 1,000 per page)
let cursor;
const allResults = [];

while (true) {
  const path = `/extractions/${job.id}${cursor ? `?after=${cursor}` : ""}`;
  const page = await xquikFetch(path);
  allResults.push(...page.results);
  // Each result: { xUserId, xUsername, xDisplayName, xFollowersCount, xVerified, xProfileImageUrl }

  if (!page.hasMore) break;
  cursor = page.nextCursor;
}

// Step 5: Export as CSV/XLSX/Markdown (50,000 row limit)
const exportUrl = `${BASE}/extractions/${job.id}/export?format=csv`;
const csvResponse = await fetch(exportUrl, { headers });
const csvData = await csvResponse.text();

Orchestrating Multiple Extractions

When building applications that combine multiple extraction tools (e.g., market research), run them sequentially and respect rate limits:

async function marketResearchPipeline(username) {
  // 1. Get user profile
  const user = await xquikFetch(`/x/users/${username}`);

  // 2. Extract their recent posts
  const postsJob = await xquikFetch("/extractions", {
    method: "POST",
    body: JSON.stringify({ toolType: "post_extractor", targetUsername: username }),
  });

  // 3. Search for related conversations
  const tweets = await xquikFetch(`/x/tweets/search?q=from:${username}`);

  // 4. For top tweets, extract replies for sentiment analysis
  for (const tweet of tweets.tweets.slice(0, 5)) {
    const estimate = await xquikFetch("/extractions/estimate", {
      method: "POST",
      body: JSON.stringify({ toolType: "reply_extractor", targetTweetId: tweet.id }),
    });

    if (estimate.allowed) {
      const repliesJob = await xquikFetch("/extractions", {
        method: "POST",
        body: JSON.stringify({ toolType: "reply_extractor", targetTweetId: tweet.id }),
      });
      // Process replies...
    }
  }

  // 5. Get trending topics for context
  const trends = await xquikFetch("/trends?woeid=1");

  return { user, posts: postsJob, tweets, trends };
}

Giveaway Draws

Run transparent, auditable giveaway draws from tweet replies with configurable filters.

Create Draw Request

POST /draws with a tweetUrl (required) and optional filters:

Field	Type	Description
tweetUrl	string	Required. Full tweet URL: https://x.com/user/status/ID
winnerCount	number	Winners to select (default 1)
backupCount	number	Backup winners to select
uniqueAuthorsOnly	boolean	Count only one entry per author
mustRetweet	boolean	Require participants to have retweeted
mustFollowUsername	string	Username participants must follow
filterMinFollowers	number	Minimum follower count
filterAccountAgeDays	number	Minimum account age in days
filterLanguage	string	Language code (e.g., "en")
requiredKeywords	string[]	Words that must appear in the reply
requiredHashtags	string[]	Hashtags that must appear (e.g., ["#giveaway"])
requiredMentions	string[]	Usernames that must be mentioned (e.g., ["@xquik"])

Complete Draw Workflow

// Step 1: Create draw with filters
const draw = await xquikFetch("/draws", {
  method: "POST",
  body: JSON.stringify({
    tweetUrl: "https://x.com/burakbayir/status/1893456789012345678",
    winnerCount: 3,
    backupCount: 2,
    uniqueAuthorsOnly: true,
    mustRetweet: true,
    mustFollowUsername: "burakbayir",
    filterMinFollowers: 50,
    filterAccountAgeDays: 30,
    filterLanguage: "en",
    requiredHashtags: ["#giveaway"],
  }),
});
// Response:
// {
//   id: "42",
//   tweetId: "1893456789012345678",
//   tweetUrl: "https://x.com/burakbayir/status/1893456789012345678",
//   tweetText: "Giveaway! RT + Follow to enter...",
//   tweetAuthorUsername: "burakbayir",
//   tweetLikeCount: 5200,
//   tweetRetweetCount: 3100,
//   tweetReplyCount: 890,
//   tweetQuoteCount: 45,
//   status: "completed",
//   totalEntries: 890,
//   validEntries: 312,
//   createdAt: "2026-02-24T10:00:00.000Z",
//   drawnAt: "2026-02-24T10:01:00.000Z"
// }

// Step 2: Get draw details with winners
const details = await xquikFetch(`/draws/${draw.id}`);
// details.winners: [
//   { position: 1, authorUsername: "winner1", tweetId: "...", isBackup: false },
//   { position: 2, authorUsername: "winner2", tweetId: "...", isBackup: false },
//   { position: 3, authorUsername: "winner3", tweetId: "...", isBackup: false },
//   { position: 4, authorUsername: "backup1", tweetId: "...", isBackup: true },
//   { position: 5, authorUsername: "backup2", tweetId: "...", isBackup: true },
// ]

// Step 3: Export results
const exportUrl = `${BASE}/draws/${draw.id}/export?format=csv`;

Webhook Event Handling

Webhooks deliver events to your HTTPS endpoint with HMAC-SHA256 signatures. Each delivery is a POST with X-Xquik-Signature header and JSON body containing eventType, username, and data.

Webhook Handler (Express)

import express from "express";
import { createHmac, timingSafeEqual, createHash } from "node:crypto";

const WEBHOOK_SECRET = process.env.XQUIK_WEBHOOK_SECRET;
const processedHashes = new Set(); // Use Redis/DB in production

function verifySignature(payload, signature, secret) {
  const expected = "sha256=" + createHmac("sha256", secret).update(payload).digest("hex");
  return timingSafeEqual(Buffer.from(expected), Buffer.from(signature));
}

const app = express();

app.post("/webhook", express.raw({ type: "application/json" }), (req, res) => {
  const signature = req.headers["x-xquik-signature"];
  const payload = req.body.toString();

  // 1. Verify HMAC signature (constant-time comparison)
  if (!signature || !verifySignature(payload, signature, WEBHOOK_SECRET)) {
    return res.status(401).send("Invalid signature");
  }

  // 2. Deduplicate (retries can deliver the same event twice)
  const payloadHash = createHash("sha256").update(payload).digest("hex");
  if (processedHashes.has(payloadHash)) {
    return res.status(200).send("Already processed");
  }
  processedHashes.add(payloadHash);

  // 3. Parse and route by event type
  const event = JSON.parse(payload);
  // event.eventType: "tweet.new" | "tweet.reply" | "tweet.quote" | "tweet.retweet" | "follower.gained" | "follower.lost"
  // event.username: monitored account username
  // event.data: tweet data ({ tweetId, text, metrics }) or follower data ({ followerId, followerUsername, followerName, followerFollowersCount, followerVerified })

  // 4. Respond within 10 seconds (process async if slow)
  res.status(200).send("OK");
});

app.listen(3000);

For Flask (Python) webhook handler, see references/python-examples.md.

Webhook security rules:

• Always verify signature before processing (constant-time comparison)
• Compute HMAC over raw body bytes, not re-serialized JSON
• Respond 200 within 10 seconds; queue slow processing for async
• Deduplicate by payload hash (retries can deliver same event twice)
• Store webhook secret in environment variables, never hardcode
• Retry policy: 5 attempts with exponential backoff on failure

Check delivery status via GET /webhooks/{id}/deliveries to monitor successful and failed attempts.

Real-Time Monitoring Setup

Complete end-to-end: create monitor, register webhook, handle events.

// 1. Create monitor
const monitor = await xquikFetch("/monitors", {
  method: "POST",
  body: JSON.stringify({
    username: "elonmusk",
    eventTypes: ["tweet.new", "tweet.reply", "tweet.quote", "follower.gained"],
  }),
});
// Response: { id: "7", username: "elonmusk", xUserId: "44196397", eventTypes: [...], createdAt: "..." }

// 2. Register webhook
const webhook = await xquikFetch("/webhooks", {
  method: "POST",
  body: JSON.stringify({
    url: "https://your-server.com/webhook",
    eventTypes: ["tweet.new", "tweet.reply"],
  }),
});
// IMPORTANT: Save webhook.secret. It is shown only once!

// 3. Poll events (alternative to webhooks)
const events = await xquikFetch("/events?monitorId=7&limit=50");
// Response: { events: [...], hasMore: false }

Event types: tweet.new, tweet.quote, tweet.reply, tweet.retweet, follower.gained, follower.lost.

MCP Server (AI Agents)

The MCP server at https://xquik.com/mcp uses a code-execution sandbox model with 2 tools (explore + xquik). The agent writes async JavaScript arrow functions that run in a sandboxed environment with auth injected automatically. StreamableHTTP transport. API key auth (x-api-key header) for CLI/IDE clients; OAuth 2.1 for web clients (Claude.ai, ChatGPT Developer Mode). Supported platforms: Claude.ai, Claude Desktop, Claude Code, ChatGPT (Custom GPT, Agents SDK, Developer Mode), Codex CLI, Cursor, VS Code, Windsurf, OpenCode.

Legacy v1 server at https://xquik.com/mcp/v1 exposes 18 discrete tools with traditional input schemas. All new integrations should use the default v2 server at /mcp.

The server also registers 5 guided workflow prompts: compose-tweet, compose-trending-tweet, compose-radar-tweet, analyze-account, run-giveaway. Use prompts/list and prompts/get in compatible clients.

For setup configs per platform, read references/mcp-setup.md. For the complete v1 tool reference with input/output schemas, annotations, and selection rules, read references/mcp-tools.md.

MCP vs REST API

	MCP Server (v2)	REST API
Best for	AI agents, IDE integrations	Custom apps, scripts, backend services
Model	2 tools (explore + xquik) with code-execution sandbox	76 individual endpoints
Categories	9: account, composition, extraction, integrations, media, monitoring, twitter, x-accounts, x-write	Same
User profile	Full (via xquik tool calling REST endpoints)	Full profile
Search results	Full (via xquik tool)	Includes optional engagement metrics
Webhook/monitor update	Full PATCH via xquik tool	PATCH endpoints
Write actions	Full via xquik tool (tweet, like, follow, DM, etc.)	POST/DELETE endpoints
File export	Not available	CSV, XLSX, Markdown
Unique to REST	-	API key management, file export (CSV/XLSX/MD), account locale update

Use the REST API GET /x/users/{username} for the complete user profile with verified, location, createdAt, and statusesCount fields.

Workflow Patterns

Common multi-step tool sequences:

• Set up real-time alerts: monitors (action=add) -> webhooks (action=add) -> webhooks (action=test)
• Run a giveaway: get-account (check budget) -> draws (action=run)
• Bulk extraction: get-account (check subscription) -> extractions (action=estimate) -> extractions (action=run) -> extractions (action=get, results)
• Full tweet analysis: lookup-tweet (metrics) -> extractions (action=run) with thread_extractor (full thread)
• Find and analyze user: get-user-info (profile) -> search-tweets from:username (recent tweets) -> lookup-tweet (metrics on specific tweet)
• Compose algorithm-optimized tweet: compose-tweet (step=compose) -> AI asks follow-ups -> compose-tweet (step=refine) -> AI drafts tweet -> compose-tweet (step=score) -> iterate
• Analyze tweet style: styles (action=analyze, fetch & cache tweets) -> styles (action=get, reference) -> compose-tweet with styleUsername
• Compare styles: styles (action=analyze) for both accounts -> styles (action=compare)
• Track tweet performance: styles (action=analyze, cache tweets) -> styles (action=analyze-performance, live metrics)
• Save & manage drafts: compose-tweet -> drafts (action=save) -> drafts (action=list) -> drafts (action=get/delete)
• Download & share media: download-media (returns permanent hosted URLs)
• Get trending news: get-radar (7 sources, free) -> compose-tweet with trending topic
• Subscribe or manage billing: subscribe (returns Stripe URL)
• Post a tweet: connect X account -> POST /x/tweets with account + text (optionally attach media via POST /x/media first)
• Engage with tweets: POST /x/tweets/{id}/like, POST /x/tweets/{id}/retweet, POST /x/users/{id}/follow
• Set up Telegram alerts: POST /integrations (type=telegram, chatId, eventTypes) -> POST /integrations/{id}/test

Pricing & Quota

• Base plan: $20/month (1 monitor, monthly usage quota)
• Extra monitors: $5/month each
• Per-operation costs: tweet search $0.003, user profile $0.0036, follower fetch $0.003, verified follower fetch $0.006, follow check $0.02, media download $0.003, article extraction $0.02
• Free: account info, monitor/webhook management, radar, extraction history, cost estimates, tweet composition (compose, refine, score), style cache management (list, get, save, delete, compare), drafts, X account management (connect, list, disconnect, reauth), integration management (create, list, update, delete, test)
• Metered: tweet search, user lookup, tweet lookup, follow check, media download (first download only, cached free), extractions, draws, style analysis, performance analysis, trends, write actions (tweet, like, retweet, follow, DM, profile, media upload, communities)
• Extra usage: enable from dashboard to continue metered calls beyond included allowance. Tiered spending limits: $5 -> $7 -> $10 -> $15 -> $25 (increases with each paid overage invoice)
• Quota enforcement: 402 usage_limit_reached when included allowance exhausted (or 402 overage_limit_reached if extra usage is active and spending limit reached)
• Check usage: GET /account returns usagePercent (0-100)

Conventions

• IDs are strings. Bigint values; treat as opaque strings, never parse as numbers
• Timestamps are ISO 8601 UTC. Example: 2026-02-24T10:30:00.000Z
• Errors return JSON. Format: { "error": "error_code" }
• Cursors are opaque. Pass nextCursor as the after query parameter, never decode
• Export formats: csv, xlsx, md via GET /extractions/{id}/export?format=csv or GET /draws/{id}/export?format=csv&type=winners

Reference Files

For additional detail beyond this guide:

• references/mcp-tools.md: All 18 legacy v1 MCP tools with input/output schemas, annotations, selection rules, workflow patterns, common mistakes, and unsupported operations
• references/api-endpoints.md: All REST API endpoints with methods, paths, parameters, and response shapes
• references/python-examples.md: Python equivalents of all JavaScript examples (retry, extraction, draw, webhook)
• references/webhooks.md: Extended webhook examples, local testing with ngrok, delivery status monitoring
• references/mcp-setup.md: MCP server configuration for 10 IDEs and AI agent platforms
• references/extractions.md: Extraction tool details, export columns
• references/types.md: TypeScript type definitions for all REST API and MCP output objects