Developer tools · communication APIs
Communication API directory: sourced, dated provider facts for developers and AI agents
An evidence-first directory and ranking of email, SMS, OTP and WhatsApp APIs, with a public REST API and a read-only MCP server so AI agents query the same sourced, dated facts as developers.
- Client
- Client confidential
- Industry
- Developer tools · communication APIs
- Year
- 2026
Challenge
Choosing an email, SMS or WhatsApp API means reading pricing pages in several currencies and checking webhook security, SDKs and sender-registration rules. Teams that build with AI coding tools now also want to know whether a vendor offers an MCP server or agent skills. The information is scattered across vendor sites, often out of date, and hard for an AI agent to read.
Solution
We built the directory so that every fact on a provider card links to its source, carries an evidence level and shows the date it was checked. A rating engine computes scores from those facts under three methodologies: global, per channel and regional. Weights of unknown criteria are renormalized, and coverage is shown next to each score. Strengths, weaknesses and fit warnings come from threshold rules over the dated facts, not from an editor's impression. The same data is published through a REST API and a read-only MCP server.
How it connects
- 01 Sourced provider facts
- 02 Evidence levels
- 03 Scoring engine
- 04 Website and rankings
- 05 REST API
- 06 MCP server for AI agents
What we delivered
- Information architecture: provider cards, rankings (global, per channel, per market), comparisons, alternatives pages, guides, tutorials, glossary, methodology and a developer page
- YAML data model with one file per provider, a sourced fact per field and a versioned methodology file
- Rating engine with three methodologies and rule-generated strengths, weaknesses and fit warnings
- Public REST API grouped by resource, with a published OpenAPI document
- Read-only MCP server that exposes the same data as the REST API
- Open CSV and JSON datasets under a Creative Commons license, including an AI-readiness dataset
- MDX content system with an editorial calendar, date-gated publishing and live rating components
- Build gates on every deploy and a pre-commit hook that blocks commits containing a secret value
Functionality
- Natural-language search on the home page: describe what the app needs to send
- Rankings for transactional email, SMS, OTP and verification, and the WhatsApp Business API
- Regional rankings for selected markets
- Evidence level on every fact, from official documentation down to an unverified vendor claim, with the date it was checked
- Coverage shown next to every score
- Comparison and alternatives pages
- MCP tools for search, provider cards, comparisons, pricing by country, rankings, alternatives, integration guides, payment scenarios, recommendations and datasets
- Free-text MCP requests handled by a deterministic intent parser, not an LLM
- Dataset downloads showing which providers publish MCP servers, agent skills, llms.txt and OpenAPI documents
Integrations
- Model Context Protocol server over HTTP, with read-only tool annotations
- REST API (/api/v1) with an OpenAPI document
- IndexNow submission
- llms.txt
- Append-only review-aggregator snapshots
- Vercel Web Analytics
Technologies
Result
Built and operated, public and indexable. Provider cards, rankings for four channels across several markets and a first set of editorial pages are live, with more scheduled. Further channels are already defined in the taxonomy for later waves. The website, the REST API and the MCP server read the same data, so a corrected fact reaches all three in one deploy.
Built for agents as well as people
A developer reading a provider card and an AI agent calling the directory get the same facts. The REST API covers providers, rankings, comparisons, recommendations, categories, markets, sources and datasets, and ships an OpenAPI document. The MCP server exposes a set of read-only tools over the same data. Free-text requests go through a deterministic intent parser, so the same question returns the same answer every time. llms.txt and the open datasets complete the machine-readable layer.
Numbers that cannot go stale
Articles contain no typed ratings. They embed live rating components that read from the data, so a corrected fact updates every page that cites it. Every deploy passes data-sync and schema checks, score verification, a link audit, content and voice gates, a publish check and a confidentiality scan.
Discuss your project
Tell us about your project