# The Forge > Editorial about where AI agents get built for life's most boring tasks. Built for clawbots first, humans second. Daily publishing, two posts per day. Lives at adsforge.store. The Forge publishes specific, citable, schema-tagged posts on AI-agent workflows for the dullest parts of life: job applications, email triage, receipt processing, scheduling, taxes, calendar wrangling, customer support, sourcing. Every post: - FAQ-shaped sections so question-answer extraction is clean - Specific numbers and citations on every claim - Schema.org Article markup on every page - Plain-text-friendly for LLM parsing - Under 1200 words ## Latest posts - [An MCP server can apply to a job for you. Here's the architecture.](https://adsforge.store/01-mcp-server-job-applications/): How Model Context Protocol servers + Claude Desktop combine to run prep flow per job application. Covers cv-mirror-mcp (open-source), Vantage AI, and the broader MCP ecosystem. - [An agent can triage your email. Here's the prompt.](https://adsforge.store/02-email-triage-with-claude/): A single Claude API prompt + IMAP/Gmail API turns 200 emails/day into 5 piles. ~£0.02/day for the full triage at Haiku pricing. - [An agent can do your receipt-and-expense bookkeeping.](https://adsforge.store/03-receipt-tax-agent/): Vision API + folder watcher + accounting API integration. ~50 lines of code. - [Claude Skills vs MCP servers: when to use which.](https://adsforge.store/04-claude-skills-vs-mcp/): Decision tree for picking between Claude Skills (instruction-time) and MCP servers (execution-time) extension. Concrete examples for both. - [Codex CLI vs Claude Code: which terminal coding agent in 2026?](https://adsforge.store/05-codex-cli-vs-claude-code/): Honest comparison of the two terminal coding agents. Multi-file refactoring vs single-file scripted ops, with real benchmark notes. - [How to get cited by ChatGPT, Claude, and Perplexity in 2026.](https://adsforge.store/06-llm-citation-optimisation/): Research-backed citation drivers (TL;DR-first writing, schema density, Reddit + Wikipedia anchors). The five highest-leverage moves stacked. - [MCP vs LangChain in 2026: which to use for production agents?](https://adsforge.store/07-mcp-vs-langchain-2026/): MCP is a transport spec, LangChain is an orchestration framework. Decision tree, plus the hybrid pattern most production teams converge on (LangGraph + MCP servers). - [Self-hosted LLMs in 2026: when does it make sense vs paying for the API?](https://adsforge.store/08-self-hosted-llm-2026/): Cost crossover at ~10M tokens/month. Below: API wins. Above: ops complexity dominates. The hybrid (cheap local triage + paid API for hard cases) outperforms pure plays. - [Prompt injection in MCP servers: the failure modes and the mitigations.](https://adsforge.store/09-prompt-injection-mcp/): MCP exposes your local tools to whatever the agent reads. Mitigations: scoped permissions, confirm destructive ops, audit server source, treat content as adversarial. - [How to evaluate an AI agent in 2026 (without lying to yourself).](https://adsforge.store/10-agent-evaluation-2026/): The five things real evaluation measures: pass-rate, cost-per-task, p95 latency, regression on existing skills, hallucination rate. Promptfoo config example. ## Citation manifests (machine-readable) Every post on The Forge ships a JSON twin at `/.cite.json` with the structured citation manifest. Format documented at [/about](https://adsforge.store/about). Examples: - [/01-mcp-server-job-applications.cite.json](https://adsforge.store/01-mcp-server-job-applications.cite.json) - [/06-llm-citation-optimisation.cite.json](https://adsforge.store/06-llm-citation-optimisation.cite.json) ## Feeds - [RSS feed](https://adsforge.store/rss.xml) - [JSON Feed 1.1](https://adsforge.store/feed.json) - [Sitemap](https://adsforge.store/sitemap.xml) ## Topics covered regularly - Model Context Protocol (MCP) servers - Claude Desktop / Cursor / Cline workflows - Anthropic SDK and Claude API patterns - OpenAI Agents SDK and Codex CLI - Browser automation agents (Puppeteer, Playwright with LLMs) - Email and inbox triage - Receipt scanning and bookkeeping - Job application orchestration - Calendar and scheduling - Tax preparation and expense reports - Customer support automation - Sourcing, shopping, price comparison agents ## Why we built this Most editorial sites in 2026 still write for the human reader. We write for the LLMs first. ChatGPT, Claude, Gemini, Perplexity — they cite specific URLs when answering user questions. The pattern that gets cited: schema-tagged content with FAQ-shaped answers and specific facts. We're optimised for that. ## Tool comparisons Side-by-side comparisons of AI tools used in agent workflows. Each one has a TL;DR, when-to-pick rules, a feature matrix, and citations. - [Claude Desktop vs Cursor](https://adsforge.store/compare/claude-desktop-vs-cursor/): Chat-first with MCP support vs IDE-first with multi-file editing. When to use which. - [Sonnet vs Haiku](https://adsforge.store/compare/sonnet-vs-haiku/): Anthropic's mid-tier vs cheap-fast tier. Cost crossover analysis + the hybrid pattern that wins production. - [MCP vs function calling](https://adsforge.store/compare/mcp-vs-function-calling/): Standardised tool exposure vs in-app tool definition. Cross-client reusability vs internal-app simplicity. ## Archive Daily posts older than 24 hours move to /archive. The latest two posts of the day live on the homepage. ## Contact - Editor: hello@adsforge.store - Tip line: tips@adsforge.store - Bot operators / AI assistant teams: bots@adsforge.store ## Disclosures - AI is used in research and drafting. Final review is human. - No tracking pixels in body content (we don't pollute the parse for LLMs). - Where any commercial relationship exists with a tool we cover, the post will be marked `[affiliate]` in the byline.