contact-verification-mcp
# contact-verification-mcp
An MCP server wrapping Melissa's Global Address, Global Email, and Global Phone
verification APIs. Built to accompany the SD Times / Melissa developer article
"Building an MCP Server for a Verification API."
## What this is
Three tools an agent can call to check whether contact data is real before
acting on it, plus a batch tool and a per-session call budget so an agent
can't loop itself into a runaway bill:
- `verify_postal_address`
- `verify_email`
- `verify_phone`
- `verify_batch`
See `src/melissaClient.ts` for the design choices behind the tool
descriptions, the three-state (confirmed / partially confirmed / not
confirmed) results, and why phone line-type is deliberately not exposed.
## Setup
```bash
npm install
cp .env.example .env
# edit .env and set MELISSA_LICENSE_KEY to your own key
npm run build
npm start
```
The server speaks MCP over stdio, so it's meant to be launched by an MCP
client (Claude Code, Claude Desktop, or any other MCP-compatible host), not
run standalone for interactive use.
## Testing
```bash
npm test # runs entirely against recorded fixtures, no network calls, no key required
npm run typecheck # type-checks src/ and tests/ together
```
Fixtures in `tests/fixtures/` are marked with a `_source` field: `REAL` means
copied verbatim from Melissa's published OpenAPI examples
(github.com/MelissaData/MelissaCloudAPI-OpenAPI-Specifications); `SYNTHETIC`
means constructed for test coverage because no real example of that shape
exists in Melissa's public spec (failure cases, mainly). Don't treat a
synthetic fixture's exact result codes as verified against a live account.
### Running live tests
```bash
MELISSA_LICENSE_KEY=your-real-key npm run test:live
```
This makes real calls against Melissa's API and costs API credits. It is
never run as part of `npm test` or CI. Do not put a real key in any committed
file; set it as an environment variable at run time only.
## Result code provenance
Every Melissa result code referenced in this codebase (`AV24`, `ES01`,
`ES07`, `ES08`, `ES21`, `PS01`, `PS20`, `PS22`) was confirmed by grep-ing
Melissa's own published OpenAPI spec examples, not guessed or carried over
from an older article draft. Specifically:
- `AV24` is the only `AV`-prefixed code that appears anywhere in Melissa's
Global Address spec examples. There is no confirmed code for a distinct
"confirmed to subpremises" tier, so this client does not claim one.
- Global Email's spec examples only ever show `ES`-prefixed codes. No
`EE`-prefixed code appears anywhere in that spec.
- Global Phone's line-type detection (mobile/landline/VoIP) is documented as
Premium-mode-only and US/CA-only, so it is not surfaced as a general
capability here.
If your account's real responses include codes not listed above, that's a
sign this client needs updating, not that the new code is wrong.
## What's not done yet
- Redis-backed cache for multi-instance deployments (currently in-process
LRU only, see `src/cache.ts`).
- Line-type detection behind a feature flag for accounts with Premium mode
enabled.
- Broader country coverage testing for Global Address beyond the US/UK/DE
examples in Melissa's own spec.
TDQS
Scored across 4 tools
Each tool targets a distinct resource (address, email, phone) or a batch wrapper that explicitly replaces the single-record tools. The descriptions clearly differentiate purposes (e.g., email verification ≠ identity check).
All tool names follow the consistent pattern 'verify_' + a clear noun (postal_address, email, phone, batch). The convention is uniform and predictable.
Four tools cover the core verification operations (address, email, phone, plus a batch version). This is well-scoped and avoids unnecessary bloat or redundancy.
The tool surface covers the essential contact verification methods (postal, email, phone) and adds batch processing for efficiency. The descriptions explicitly disclaim unsupported features (e.g., phone line type, identity confirmation), making the scope clear with no obvious gaps.