# 157 MCP Tools: How We Built the Largest Agent-Callable Surface in Martech
- **Status**: published | **Author**: Lena Hartmann | **Published**: 2026-05-04T11:05:39.980542+00:00
- **Tags**: product, mcp, architecture
- **Reading time**: 2 min
- **Views**: 478

## Excerpt
A deep-dive into the architecture behind SpiderIQ's 157 MCP tools — schema layer, multi-tenant isolation, and the bet on Model Context Protocol.

When we launched SpiderIQ, we had 12 tools. Today, we ship 157 MCP tools across 5 dedicated servers — the largest agent-callable surface in martech. Here's how we got here, and why MCP is the right bet.

## Why Model Context Protocol?

Most automation platforms expose REST APIs. Agents can call them, sure — but every integration requires bespoke glue code: auth wrappers, pagination handling, error mapping. MCP eliminates this entirely by providing a standardized tool-calling interface that any LLM agent understands natively.

We evaluated GraphQL federation, gRPC, and OpenAPI-based approaches. MCP won because it optimizes for the agent's mental model, not the developer's. A tool is a function with typed parameters and a structured response — nothing more.

## The Five-Server Architecture

Our 157 tools are distributed across five MCP servers, each owning a bounded domain:

• SpiderMaps (42 tools) — Location intelligence, POI scraping, geo-enrichment

• SpiderVerify (31 tools) — Email/phone validation, deliverability scoring

• SpiderMail (28 tools) — Campaign orchestration, template management, sending

• SpiderCRM (34 tools) — Contact management, deal pipeline, activity tracking

• SpiderAdmin (22 tools) — Billing, usage analytics, team management

## Multi-Tenant Isolation

Every MCP call passes through our isolation layer. The agent's session token is scoped to a specific tenant, and tool execution happens inside a sandboxed worker pool. No tenant can ever read another tenant's data — even if the agent hallucinates a cross-tenant ID.

### Schema Validation

Each tool's input schema is validated at the gateway level before execution begins. We use JSON Schema draft-2020 with strict mode enabled, rejecting any undeclared properties. This catches malformed agent calls before they hit business logic.

## What's Next

We're working on tool composition — letting agents chain multiple tools into atomic transactions. Think: "scrape 50 restaurants, verify all emails, create a campaign, and schedule it" as a single, rollback-safe operation.

If you're building agent infrastructure, we'd love to compare notes. Reach out at engineering@spideriq.ai.