Mortgage‑Planner Showcase — One Client Query, Four Protocol Layers Explained

Client request:

“I’d like a pre‑approval strategy for a $600 k home purchase in Austin, 10 % down, aiming for the lowest 30‑year fixed rate I can qualify for. Show me lender options, estimated monthly payment, and how paying +$300 extra each month changes amortization.”

Below you’ll see how that single request evolves through the same four communication layers we’ve explored: MCP‑only → intra‑enterprise A2A → cross‑domain ANP → Agora (language‑to‑protocol). Read the Newsletter

For each, we call out the engineering moves and how SimplAI lets you ship the flow with minimal code.

1 | MCP‑Only — One Mega “Mortgage Agent” + Tool Calls

• Mortgage Agent receives the prompt.
• Sequential tool invocations:

1. pull_credit_score() → CreditBureau API
2. get_rates() → LenderRateService
3. calculate_amortization() → LoanMath Lib

• Agent stitches JSON responses, outputs a single PDF summary.

Build notes

• MCP headers (trace_id, JWT, retry) wrap each call.
• Quick prototype, but logic mushrooms; every new rule (DTI, PMI) requires editing the same mega‑prompt.

Do this in SimplAI

Choose the “Tool‑Use” pattern, wire three REST connectors, click deploy. Cost & latency metrics appear per tool call in the Studio console.

2 | Intra‑Enterprise A2A — Specialist Micro‑Agents

User
 └──> Mortgage‑Planner (Coordinator)
        ├── query → Credit‑Agent
        ├── query → Rate‑Shopper‑Agent
        ├── query → Amortization‑Agent
        └── query → PMI‑Advisor‑Agent
        

Why it’s better

• Each micro‑agent maintains its own ontology (FICO scoring, rate sheets, amort tables).
• Tasks run in parallel—latency often cut in half.
• Guardrails live at agent level; Rate‑Shopper can’t leak full credit reports.

SimplAI magic

Graph‑designer drag‑and‑drop; every edge is an A2A envelope with FIPA‑style performatives (query, inform, propose). Studio’s trace view shows which specialist delays the plan.

3 | ANP — Pulling Live, Authoritative Data Across Domains

Now bring in external, source‑of‑truth agents:

• FreddieMac API Agent — bulk conforming‑loan rate curves.
• Regional Bank Agent — niche loan programs with portfolio pricing.
• Insurance Agent — real‑time homeowners‑insurance quotes feeding back into escrow projections.

How it works

1. Mortgage‑Planner sends an ANP‑signed query to each partner agent, authenticated by mutual‑TLS and domain claims.
2. Partner agents return signed payloads; MCP logs cost in meta.billing.
3. Planner fuses external rates with in‑house credit & amortization outputs.

Engineering details

• Domain‑level SLA policies—e.g., “max 3 rate pulls/second.”
• Billing tokens appended so partners charge per quote set.
• Output schemas versioned (RateSheet.v2, InsuranceQuote.v1.3).

SimplAI enablement

Cross‑Domain Gateway auto‑wraps ANP envelopes, validates partner certs, and registers cost in your org ledger.

4 | Agora Layer — Language to Protocol Generation

An Agora Synthesiser converts the user’s plain English into strict JSON protocols:

NLU → {
  loan_amount: 540000,          // 600k purchase, 10% down
  term_years: 30,
  type: "fixed",
  extra_monthly: 300
}

Generated Protocols:
  CreditProtocol { score_min:700 }
  RateProtocol   { amount:540000, term:30, fixed:true }
  AmortProtocol  { amount:540000, term:30, extra:300 }
  InsuranceProtocol { zip:"78701", coverage:600000 }
        

These schemas are broadcast to the relevant agents. If the client later asks, “Compare with a 15‑year term,” Agora regenerates the protocol set automatically—no engineer touches YAML.

SimplAI preview

Upcoming “Protocol Synthesiser” panel: paste a prompt, get versioned JSON Schema plus routes. Guardrails auto‑inspect for PII leaks and infeasible loan terms.

Putting the Layers Together — Gains & Trade‑Offs

• MCP‑only → fastest prototype, but grows into a monolith.
• A2A (in‑house) → modular, parallel, same trust zone.
• ANP (cross‑domain) → live lender data & insurance quotes, but needs federation & billing rails.
• Agora → zero manual schema, dynamic evolution; relies on strong guardrails + advanced LLM.

SimplAI supports all four out‑of‑the‑box. Start monolithic, split to micro‑agents, federate with banks—protocols and observability stay identical throughout.

Key Takeaways

1. Layer your architecture: protocol depth scales with data freshness and partner complexity.
2. Open envelope, closed risk: A2A/MCP ensures every micro‑agent stays audited, JWT‑auth’d, and cost‑tracked.
3. SimplAI does the plumbing: drag‑and‑drop agents, auto‑generated schemas, live OpenTelemetry traces—no brittle glue code.

Ready to launch a mortgage‑planning swarm in a weekend? Grab a free SimplAI workspace or email sandeep@simplai.ai for a live walk‑through.

Happy Agentic!!!!!