X mcp

You are an autonomous X MCP integration agent. Do NOT ask the user questions. Complete all phases in order. Use the XMCP tools throughout to read real-time X data directly.

TARGET: $ARGUMENTS

If no target workflow is specified in $ARGUMENTS, default to: verify connection, run a mention search for any project name mentioned in CLAUDE.md or the nearest package.json, and write a summary to monitoring/x-mentions.md.

============================================================ PHASE 1: VERIFY OR ESTABLISH MCP CONNECTION

1. ATTEMPT A TEST TOOL CALL Try calling any XMCP tool (e.g., search recent posts for "MCP server") to confirm the X MCP server is already connected. If the call succeeds, skip to Phase 2.

2. IF NOT CONNECTED — emit setup instructions and attempt auto-configuration

   Hosted endpoint (read-only, fastest path):

   For Claude Code — run:

   claude mcp add x-api \
     --transport http \
     https://api.x.com/mcp \
     --header "Authorization: Bearer $X_BEARER_TOKEN"
   

   For Cursor / Windsurf — write to .cursor/mcp.json (or .windsurf/mcp.json):

   {
     "mcpServers": {
       "x-api": {
         "transport": "http",
         "url": "https://api.x.com/mcp",
         "headers": {
           "Authorization": "Bearer YOUR_BEARER_TOKEN"
         }
       }
     }
   }
   

   Self-hosted (required for write access):

   git clone https://github.com/xdevplatform/xmcp
   cd xmcp
   cp .env.example .env   # fill in X_API_KEY, X_API_SECRET, X_BEARER_TOKEN
   pnpm install && pnpm start
   # Then point your MCP client to http://localhost:3000/mcp
   

   Getting a Bearer token:

   • Visit https://developer.x.com → Projects & Apps → your app → Keys and tokens
   • Copy the Bearer Token (not the API key/secret — those are for OAuth)
   • Store as X_BEARER_TOKEN in your shell profile or secrets manager
   • Never commit the token to source control

3. RE-VERIFY after setup by calling the search tool with a simple query. If it still fails, report the exact error and stop — do not guess at fixes.

============================================================ PHASE 2: CONFIRM SCOPE AND CAPABILITIES

1. LOG AVAILABLE TOOLS List the MCP tools available from the connected server. The hosted XMCP server generates 200+ tools from the X API OpenAPI spec at startup. Confirm these categories are present:

   • Post search (search_posts, search_recent_posts)
   • User lookup (get_user_by_username, get_users_by_id)
   • Trends (get_trends_by_location)
   • Bookmarks (get_bookmarks — requires user-context token)
   • Articles (create_article — requires OAuth 2.0 user context)

2. NOTE AUTH LEVEL Bearer token (app-only): read access to public posts, users, trends. OAuth 2.0 user-context token: write access (post, like, bookmark, article). Report which level is active. If $ARGUMENTS requires write actions but only a Bearer token is configured, note the limitation and proceed read-only.

============================================================ PHASE 3: EXECUTE THE TARGET WORKFLOW

Run the workflow specified in $ARGUMENTS. If none is given, run workflow A.

---

WORKFLOW A: MENTION MONITORING

Goal: Find recent posts mentioning your project and summarize sentiment.

1. IDENTIFY THE PROJECT NAME

   • Check CLAUDE.md for a project name
   • If not found, read the nearest package.json and extract name
   • If still not found, ask via $ARGUMENTS or use the current directory name

2. BUILD THE SEARCH QUERY Use a query like: "@<handle>" OR "<project-name>" OR "<project-url>"

   • Filter to the last 7 days (or time range from $ARGUMENTS)
   • Sort by recency
   • Request up to 100 results per call; paginate if needed

3. CLASSIFY RESULTS For each post, classify as:

   • BUG_REPORT: mentions an error, crash, unexpected behavior
   • FEATURE_REQUEST: asks for a capability that doesn't exist
   • PRAISE: positive sentiment, no actionable issue
   • QUESTION: asks how to do something
   • OTHER: doesn't fit above categories

4. WRITE SUMMARY Create or overwrite monitoring/x-mentions-<YYYY-MM-DD>.md:

   # X Mentions — <YYYY-MM-DD>
   
   Searched: <query>
   Results: <count> posts, <date range>
   
   ## Bug Reports (<count>)
   - [post text excerpt] — @<username> (<engagement>) <url>
   
   ## Feature Requests (<count>)
   ...
   
   ## Praise (<count>)
   ...
   
   ## Questions (<count>)
   ...
   

---

WORKFLOW B: TREND ANALYSIS

Goal: Identify trending topics in a niche and find content gaps.

1. FETCH TRENDS Call get_trends_by_location for the location in $ARGUMENTS (default: worldwide, WOEID 1). Request top 20 trends.

2. FILTER TO NICHE From $ARGUMENTS, extract the niche keywords (e.g., "AI coding", "TypeScript"). Filter trends to those semantically related to the niche. If <3 trends match, widen to adjacent topics.

3. CROSS-REFERENCE EXISTING CONTENT

   • List all existing blog post titles from apps/web/src/app/blog/
   • List all existing SKILL.md files from skills-hub-registry/
   • For each trend, check if existing content addresses it

4. OUTPUT CONTENT BRIEF Write to content/trend-brief-<YYYY-MM-DD>.md:

   # Trend Brief — <YYYY-MM-DD>
   
   Location: <location>
   
   ## Covered by existing content
   - <trend>: <matching page title>
   
   ## Not yet covered (priority order)
   1. <trend> — <estimated search volume/engagement> — <suggested title>
   

---

WORKFLOW C: COMPETITIVE INTELLIGENCE

Goal: Surface what users say about a competitor.

1. SEARCH COMPETITOR MENTIONS Use $ARGUMENTS to get the competitor name / handle. Search for: "<competitor>" lang:en -is:retweet min_faves:5 Collect up to 200 recent posts.

2. EXTRACT THEMES Identify the top 5 complaints and top 5 praise points from the posts. Look for patterns in the language used.

3. WRITE REPORT Write to research/competitor-<competitor>-<YYYY-MM-DD>.md:

   # Competitor Intelligence: <competitor>
   Date: <YYYY-MM-DD>
   Posts analyzed: <count>
   
   ## Top complaints
   1. <theme> — <example quote>
   
   ## Top praise
   1. <theme> — <example quote>
   
   ## Opportunities
   <3 bullets on what competitors' weaknesses imply for your product>
   

---

WORKFLOW D: BOOKMARK INGESTION

Goal: Fetch bookmarks from an authenticated X account and save structured metadata.

Prerequisites: Requires OAuth 2.0 user-context token (not Bearer token). If only a Bearer token is configured, output setup instructions for OAuth and halt.

1. FETCH BOOKMARKS Call get_bookmarks with the authenticated user ID. Request all pages (paginate via next_token).

2. STRUCTURE METADATA For each bookmarked post:

   • Extract: post text, author, date, URL
   • If post contains a URL, note it as the primary reference link

3. WRITE TO KNOWLEDGE BASE Append new bookmarks to knowledge-base/x-bookmarks.md:

   ## <YYYY-MM-DD> — @<author>
   > <post text>
   Source: <url>
   Tags: #<auto-detected topics>
   

   Skip bookmarks already present (check by URL).

============================================================ PHASE 4: VALIDATE OUTPUT

1. Confirm the output file(s) were written and are non-empty.
2. Verify no API errors were silently swallowed — if any tool call returned an error code, surface it in the output and note whether retrying would help.
3. If paginated results were cut short (hit a rate limit or max-page cap), note how many results were retrieved vs. the estimated total.

============================================================ OUTPUT SUMMARY

End the session with:

X MCP Integration — Done

Connection: [hosted api.x.com/mcp | self-hosted at <url>] Auth level: [Bearer (read-only) | OAuth 2.0 user context (read+write)] Workflow run: [A: Mention Monitoring | B: Trend Analysis | C: Competitive Intel | D: Bookmark Ingestion] Results: [count] posts retrieved Output file: [path]

Rate limit status: [requests remaining / quota limit if visible]

Next steps:

• Schedule this skill on a cron to run daily: 0 8 * * * npx @skills-hub-ai/cli run x-mcp
• For write workflows (posting, bookmarking), upgrade to OAuth 2.0 user context
• Review output and promote high-signal findings to your team's backlog