Ask HN: Who is using MCP in production?

Hacker News Discussion ↗

Developers are increasingly deploying Model Context Protocol (MCP) servers in production to let AI agents safely interact with enterprise systems like Jira, GitHub, CRMs, monitoring tools, and custom internal apps. Supporters say MCP standardizes tool discovery, OAuth-based auth, and fine-grained capability boundaries—especially valuable for non-technical users and third-party voice/chat platforms—while critics argue many use cases are better served by direct APIs or CLIs as models improve at tool use. Overall, MCP is emerging as a de facto integration layer for AI agents in multi-system, multi-vendor environments, even as its long-term necessity and efficiency remain contested.

Overall sentiment

  • Many commenters are actively using MCP (Model Context Protocol) in production; others are skeptical or think its value is shrinking as models and CLIs improve.
  • Strong divide between “agent-in-terminal for developers” vs “end‑user / enterprise integrations,” where MCP is seen as much more useful.

Common production use cases

  • Customer‑facing and internal voice agents: appointment scheduling, order status, CRM access, phone-based receptionists.
  • SaaS products: CRM/ERP, hosting/control panels, monitoring, compliance tools, reporting/analytics, social media management, help centers, e‑commerce analytics, niche vertical tools.
  • Developer workflows: bug/feature reporting, task tracking, project management (Jira/Linear/etc.), code debugging across logs/metrics/alerts, infra provisioning.
  • Personal/experimental tools: learning assistants, custom data archives, grocery shopping, language study, home automation, visual effects workflows, personal ticket boards.
  • Cross‑system “gateway” MCPs that unify logs, metrics, multiple APIs, or legacy systems into a single agent‑friendly surface.

Perceived advantages

  • Standardized, agent‑native interface:
    • Tools/resources are self‑describing and discoverable.
    • Easier to expose “recipes” / workflows than raw API endpoints.
  • Auth and security:
    • OAuth + standardized flows; users connect via a click/OAuth instead of managing API keys.
    • Can tightly scope capabilities and hide underlying service credentials from the model.
  • Distribution & usability:
    • Non‑technical users can connect products into ChatGPT/Claude/etc. without CLIs.
    • Single MCP often reused by in‑app assistants, external agents, and multiple vendors.
  • Abstraction:
    • Wraps ugly, inconsistent, or hard‑to-change REST/CLI surfaces with a cleaner, LLM‑tuned layer.

Critiques and skepticism

  • Many developers prefer CLIs or direct APIs:
    • Cheaper, more efficient token use; easier to debug; less context bloat.
    • Skills + shell access often sufficient.
  • Complaints that MCP servers vary widely in quality; many are worse than direct APIs.
  • Some see MCP as a fad or unnecessary abstraction now that models handle APIs/CLIs better.
  • Building a “well‑engineered” MCP is described as heavy: requires careful design, benchmarking, and ongoing iteration.
  • Questions about marginal benefit over “good REST + docs,” especially for dev-heavy teams.

Nuances & open questions

  • Token efficiency and performance comparisons are contested.
  • Lazy loading of tools and tool search can mitigate overhead but depend on the harness.
  • Long‑term role of MCP vs. APIs/skills/CLIs is debated and unresolved.