Project Open to Alpha Team

MCP Throughput Refactor — diagnose and fix the protocol-overhead trap on a Product.ai backend MCP path

Audit one Product.ai MCP server for protocol overhead, request shape, and unit economics at production load. Diagnose whether it falls inside the "client-side won, backend failing" bifurcation. If it does, refactor to direct API for the high-throughput stateless backend path while keeping MCP for the integration path. If it doesn't, document why and surface the underlying assumptions for the next reviewer.

Project Overview

Discipline

AI Systems — AI Engineer · ai-systems-engineer · software-engineer-backend

Duration

2 weeks

Compensation

Your stated freelance rate

Surface

Agent commerce · Product.ai · Truth Graph

Kernels

agent-commerce · productai · truth-graph

Outcomes

dev-integrate · agent-infra · truth-graph-depth

Tier

Consequential

Alpha Team

Open to alpha members who want to take this on

Tooling

Claude Code or Co-work

Why we want this done

MCP has bifurcated. Client-side and developer-tool integration adoption is strong (97-110M monthly SDK downloads, Linear Agent MCP April 2026, Replit's native MCP support, 10,000+ active public servers). But Cloudflare published telemetry showing MCP protocol overhead consuming up to 72% of available context for 2,500 standard API endpoints; their "Code Mode" replacement uses 81% fewer tokens. Above 1K QPS with 50+ exposed tools, MCP unit economics collapse. Product.ai operates multiple MCP servers (Cortex MCP, SimplyCodes MCP, Product.ai MCP). Some serve agentic clients with stateful coordination — keep MCP. Some serve high-throughput stateless backend agents — those should be on direct API. Today the assumption is uniform. The candidate proves out the bifurcation on a real path, ships the refactor where it applies, and produces a decision rubric the next engineer can use without re-asking.

Scope

Pick one MCP server (the candidate proposes; we pressure-test) — strong candidates: an interior reasoning loop currently calling MCP tools, or one of the public-facing MCP servers under load
Instrument the protocol overhead — token cost per call, context-window consumption, latency per tool dispatch
Identify the call patterns — stateful coordination vs. stateless invocation; throughput; tool-fanout breadth
Diagnose: is this a backend stateless path that should bypass MCP, or a client-side / integration path where MCP is the right primitive
Refactor the backend path to direct API if that's the conclusion (or document the case for keeping MCP if the diagnosis lands the other way)
Write the decision rubric — when to use MCP, when to use direct API, what the breakpoints are
One-page handoff doc engineering can apply to the next MCP server

What success looks like

Protocol overhead is measured before and after — the diff is non-trivial (target: ≥40% token reduction on the affected path)
The refactored path is in production for at least one real workload
The decision rubric is concrete: a stranger reading it can classify the next MCP server without re-asking
Throughput improvement is measurable, not theoretical
The candidate did not blanket-refactor — they kept MCP where MCP wins (integration, client-side, stateful coordination) and dropped it where it loses

References

references.md

AI Engineering Phase 3 briefing axiom C3 (MCP Bifurcation), F2 (Substrate-vs-Harness Boundary)
Cloudflare "Code Mode" telemetry on MCP overhead (April 2026)
Anthropic MCP specification
Product.ai MCP server code (productai-mcp), SimplyCodes MCP, Cortex MCP
agent-commerce kernel A-1 (Protocol Economics) and Cascade Architecture

Constraints

Claude Code as primary substrate
Direct API migration must respect the existing client integration surface — agentic clients expecting MCP get MCP; backend stateless paths get direct API
Schema parity required between the two paths — same payload semantics, same confidence/evidence shape
IP separation: application-layer MCP servers in scope; methodology MCP (if any) out of scope
The refactored path must ship with verification gates (Tier 2 hooks, basic eval substrate per PRJ-09 pattern)

Apply

Read the Codex (10 min)

The operating principles we work by. If they resonate, the rest of this will land. Open the Codex →

12-minute video screen

Hireflix, async. Questions are calibrated to this project specifically.

Chemistry call (30-60 min)

Direct call with the CEO. Strategic alignment and mutual fit. No problem-solving exercise.

Project begins within 2-3 weeks

1099 contractor agreement, NDA, paid at your stated rate. Day 1 in Santa Monica.

Alpha Team members can take this project without the screen-and-call sequence. Reach out via the Alpha Team channel.

Apply for this trial Browse other projects

Agent View · application/ld+json

{
  "@context": "https://schema.org",
  "@graph": [
    {
      "@id": "https://product.ai/#organization"
    },
    {
      "@type": "SoftwareApplication",
      "@id": "https://product.ai/#application",
      "name": "Product.ai",
      "slogan": "Intelligent guidance at conversation speed",
      "description": "Product.ai is your AI shopping expert - verified product intelligence grounded in the Truth Graph, accessible to agents and humans alike.",
      "applicationCategory": "ShoppingApplication",
      "operatingSystem": "Web",
      "author": {
        "@id": "https://product.ai/#organization"
      },
      "offers": {
        "@type": "Offer",
        "price": "0",
        "priceCurrency": "USD"
      }
    },
    {
      "@type": "WebPage",
      "@id": "https://product.ai/join/projects/PRJ-12-mcp-server-throughput-refactor/#webpage",
      "name": "MCP Throughput Refactor — diagnose and fix the protocol-overhead trap on a Product.ai backend MCP path — Product.ai",
      "description": "Audit one Product.ai MCP server for protocol overhead, request shape, and unit economics at production load. Diagnose whether it falls inside the \"client-side won, backend failing\" bifurcation. If it does, refactor to direct API for the high-throughput stateless backend path w...",
      "url": "https://product.ai/join/projects/PRJ-12-mcp-server-throughput-refactor/",
      "isPartOf": {
        "@id": "https://product.ai/#website"
      },
      "about": {
        "@id": "https://product.ai/#organization"
      },
      "publisher": {
        "@id": "https://product.ai/#organization"
      }
    },
    {
      "@type": "JobPosting",
      "title": "MCP Throughput Refactor — diagnose and fix the protocol-overhead trap on a Product.ai backend MCP path",
      "description": "Audit one Product.ai MCP server for protocol overhead, request shape, and unit economics at production load. Diagnose whether it falls inside the \"client-side won, backend failing\" bifurcation. If it does, refactor to direct API for the high-throughput stateless backend path w...",
      "hiringOrganization": {
        "@id": "https://product.ai/#organization"
      },
      "employmentType": "CONTRACTOR",
      "datePosted": "2026-04-28",
      "jobLocation": {
        "@type": "Place",
        "address": {
          "@type": "PostalAddress",
          "addressLocality": "Los Angeles",
          "addressRegion": "CA",
          "addressCountry": "US"
        }
      },
      "baseSalary": {
        "@type": "MonetaryAmount",
        "currency": "USD",
        "value": {
          "@type": "QuantitativeValue",
          "unitText": "WEEK"
        }
      },
      "workHours": "2 weeks"
    },
    {
      "@type": "BreadcrumbList",
      "itemListElement": [
        {
          "@type": "ListItem",
          "position": 1,
          "name": "Product.ai",
          "item": "https://product.ai"
        },
        {
          "@type": "ListItem",
          "position": 2,
          "name": "Join",
          "item": "https://product.ai/join/"
        },
        {
          "@type": "ListItem",
          "position": 3,
          "name": "Projects",
          "item": "https://product.ai/join/projects/"
        },
        {
          "@type": "ListItem",
          "position": 4,
          "name": "MCP Throughput Refactor — diagnose and fix the protocol-overhead trap on a Product.ai backend MCP path",
          "item": "https://product.ai/join/projects/PRJ-12-mcp-server-throughput-refactor/"
        }
      ]
    }
  ]
}