Advanced Usage

This guide covers power-user workflows, tool chaining patterns, automation recipes, and production-scale strategies for the BPAI MCP Server.


Tool Chaining

The real power of MCP is chaining multiple tools in a single conversational flow. The AI client handles the orchestration while you describe the end goal.

Recipe 1: Generate → Publish → Index

Research runs automatically. The full content pipeline in one request:

Generate a 3000-word article about "best accounting software for freelancers 2026" 
using claude-opus-4.5 with humanize enabled,
publish it to WordPress as a draft in the "Reviews" category,
then broadcast the URL and submit to IndexNow

The AI chains: generate_article (auto-research + humanize) → publish_to_wordpressbroadcast_urlsindexnow_submit

Recipe 2: CSV → Batch Generate → Export

Bulk content production from a data file (research runs per-keyword automatically):

Import keywords from /Users/arielle/seo/q1-keywords.csv,
generate all 200 articles using gpt-5.2,
then export everything as markdown files to /Users/arielle/exports/q1/

The AI chains: import_csvgenerate_batch (auto-research per keyword) → export_markdown

Recipe 3: Multi-Project Content Sprint

Generate content across multiple projects:

Switch to my Tech Blog and generate 10 articles about cloud computing.
Then switch to my Finance Blog and generate 10 articles about personal budgeting.
Export both projects as CSV to /Users/arielle/weekly-content/

Recipe 4: Competitive Research → Content Gap Fill

Fetch the sitemap from https://competitor.com/blog-sitemap.xml,
identify topic gaps compared to my existing content,
then generate articles for the top 20 missing topics

The AI chains: fetch_sitemap (competitor) → fetch_sitemap (your site) → analysis → generate_batch

Recipe 5: Article with Hero Image → WordPress

Generate an article about "best accounting software 2026" using gemini-3.1-pro-preview.
Then generate a hero image for the same topic.
Publish the article to WordPress with the image.

The AI chains: generate_articlegenerate_imagepublish_to_wordpress

Recipe 6: Batch Infographics

For each of these 5 topics, generate an infographic image in 16:9:
- SEO growth strategies
- content marketing funnel
- social media ROI metrics
- email open rate benchmarks
- keyword research process

The AI chains: generate_image (×5) with style: infographic


CSV-Based Batch Workflows

Generate content at scale using CSV keyword imports combined with generate_batch:

Local Services Pages

Import keywords from /Users/arielle/seo/local-services.csv,
then generate all 50 articles using gpt-5.2 with word count 1500

Product Comparisons

Import keywords from /Users/arielle/seo/comparisons.csv,
generate articles using claude-opus-4.5 with word count 2500 and research

Glossary Pages

Import keywords from /Users/arielle/seo/glossary-terms.csv,
generate all with gpt-5-mini at 800 words

Article Recovery

All generated articles are automatically persisted to your article library. If a batch completes but you lose the response, you can recover:

List all articles generated in the last 24 hours
Search my article library for "accounting software"
Get the full content of article [ID]

The AI uses: list_articlesget_article


Automation Patterns

Scheduled Content Calendar

Use MCP from a script or cron job to automate content production:

#!/bin/bash
# Weekly content generation script
# Runs every Monday at 6 AM

# Generate this week's articles
claude --mcp "Import keywords from /data/weekly-keywords.csv \
  and generate all articles using gpt-5.2 with research, \
  publish to WordPress as drafts, \
  then broadcast and submit to IndexNow"

Multi-Site Publishing

Publish the same content across different WordPress sites with variations:

Generate an article about "email marketing best practices".
Publish to my Tech Blog WordPress as a draft.
Then switch to my Marketing Blog project and publish the same article there.

Content Refresh Pipeline

Update existing content with fresh research:

Export all articles from my Tech Blog older than 90 days as JSON.
For each article, run fresh research on the original keyword.
Regenerate each article with the updated research and compare word counts.
Export the updated versions as markdown.

Error Handling Strategies

Retry Logic

The MCP server implements automatic retry for transient errors:

Error Type Auto-Retry Delay Notes
Rate limited (429) Yes, 3× Exponential backoff Waits 5s, 15s, 45s
Timeout (504/408) Yes, 1× 10s Retries with same params
Server error (500) Yes, 1× 5s Provider-side issue
Auth error (401) No Check API key
Content filtered No Adjust topic
Invalid model No Check model ID

Partial Batch Failure

If some articles in a batch fail, the batch continues processing other items. Failed items are logged with error details. You can:

  1. View failures: Check the batch response results array for items with status: "failed"
  2. Retry failures: Extract the failed keywords and run generate_batch again with just those keywords
  3. Recover successes: Use list_articles to find all persisted articles from the batch

Graceful Degradation

If your primary model is unavailable, the MCP server can fall back:

Generate articles using claude-opus-4.5, fall back to gpt-5.2 if Claude 
is rate limited or unavailable

Performance Optimization

Optimal Batch Sizes

Use Case Recommended Batch Size Est. Time
Quick test 5–10 articles 5 minutes
Weekly content 20–50 articles 15–30 min
Content sprint 50–100 articles 30–60 min
Full build 100–200 articles 1–2 hours

Cost Optimization

Strategy Cost Impact Quality Impact
Use gpt-5-mini instead of gpt-5.2 ~70% cheaper Slight quality reduction
Use gpt-4.1-nano for short content ~90% cheaper Good for glossary/FAQ content
Research only high-value keywords 50% research cost saving Focus depth where it matters
Set word count to actual need Proportional savings Match content depth to intent

Token Usage

Monitor cost through your AI provider dashboards (OpenAI Usage, Anthropic Console, etc.). The MCP server logs model and provider info for each generation.


---

## Integration Examples

### Hugo Static Site

Generate 50 articles about Go programming. Export as markdown with frontmatter to /Users/arielle/hugo-site/content/posts/ Use title for filenames, include tags and date in frontmatter


The exported files are ready for `hugo build` with no additional processing.

### Astro Content Collections

Export articles as markdown to /Users/arielle/astro-site/src/content/blog/ Include all frontmatter fields: title, description, date, tags, draft


### Next.js MDX

Export articles as markdown to /Users/arielle/nextjs-site/content/ Rename .md to .mdx Include frontmatter compatible with contentlayer


### Shopify Blog

Export all articles as CSV with columns: title, body_html, handle, tags Save to /Users/arielle/shopify-import.csv


Direct Shopify CSV import support.

### Notion

Export all articles as markdown to /Users/arielle/notion-import/ Use flat structure with title as filename


Drag and drop into Notion.

---

## CLI Quick Reference

| What You Want | What to Say |
|---|---|
| Generate 1 article | "Generate an article about [topic]" |
| Generate without research | "Generate an article about [topic] with auto_research disabled" |
| Full pipeline | "Generate an article about [topic] with humanize enabled" |
| Batch from CSV | "Import keywords from [path] and generate all" |
| Publish to WordPress | "Publish article [id] to WordPress" |
| Export as markdown | "Export all articles as markdown to [path]" |
| Export as CSV | "Export project as CSV to [path]" |
| Broadcast URLs | "Broadcast URLs from batch [id]" |
| Switch project | "Switch to [project name]" |
| Check batch status | "What's the status of my batch?" |
| Generate blog images | "Generate a hero image for [topic]" |
| Infographic per post | "Generate an infographic about [topic]" |
| List image models | "What image models are available?" |

---

## Limits and Fair Use

| Resource | Free | Starter | Pro |
|---|---|---|---|
| MCP articles/month | 25 | 500 | Unlimited |
| Concurrent batches | 1 | 3 | 10 |
| Max batch size | 5 | 50 | 200 |
| CSV import rows | 50 | 5,000 | Unlimited |
| Knowledge base items | 5 | 50 | Unlimited |
| Projects | 3 | 10 | Unlimited |
| Research per month | — | 100 | Unlimited |

---

## Getting Help

- **Documentation**: [bulkpublishing.ai/docs/mcp](https://bulkpublishing.ai/docs/mcp)
- **Discord**: [discord.gg/bulkpublishing](https://discord.gg/bulkpublishing)
- **GitHub Issues**: [github.com/bulkpublishing/mcp-server/issues](https://github.com/bulkpublishing/mcp-server/issues)
- **Email**: support@bulkpublishing.ai

---

## Progress Heartbeat (v1.3.0)

The MCP server sends progress notifications every 10 seconds during long-running operations (article generation with Gemini 3.1 Pro can take 60-90 seconds). This keeps the stdio connection alive and prevents client-side timeouts.

The heartbeat is automatic. No configuration needed. Your MCP client will show progress messages like:

Generating article with gemini-3.1-pro-preview... (20s elapsed) Generating article with gemini-3.1-pro-preview... (30s elapsed)


This applies to all generation and image tools.