SmartBot platform
Slack and web platform across 150+ channels and 30+ time zones. 250+ commands and scheduled routines for 800+ employees (560k+ interactions).
Mario Ruiz Sánchez · Reykjavík
I build things: AI tools, MCP servers and experiments you can try.
Open to roles in Reykjavík & remote
I specialize in AI agents and MCP for production teams: public MCP servers, the same structured payloads in Slack and IDEs, and API tests from OpenAPI. Under that sits years of backend work, APIs, and internal platforms - SmartBot for 800+ employees, 560k+ interactions, and open-source test tooling with 3.5M+ gem downloads.
SmartBot scale across engineering. Gem downloads as of Sep 2026.
I ship through agents (Cursor, Claude Code, Claude Desktop, and other MCP clients, plus MCP servers I built): I design, review and operate. The implementation loop is part of the product.
Based in Reykjavík since 2005; also lived in Madrid, Paris, and New York.
Public read-only servers for Cursor, Claude Code, and Claude Desktop. Setup and tool reference live on the MCP catalog.
Streamable HTTP tools for CV facts, systems, blog posts, and role matching with citations.
36 tools for Icelandic earthquakes, volcanoes, weather, roads, tides, ferries, flights, buses, ski areas, trails, hot springs, events, cinema, restaurants, museums, fuel, EV chargers, parking, pharmacies, outages, and waste collection.
Slack and web platform across 150+ channels and 30+ time zones. 250+ commands and scheduled routines for 800+ employees (560k+ interactions).
Cleanup service run in Slack, web UI, REST, and dry-run to delete stale Azure test resources. It contributed to $2M+ in cloud cost reduction.
Team ChatGPT sessions with a team knowledge base as context and natural language for 50+ commands. 40k+ LLM session calls.
Same ANF analyzers as JSON API and MCP for Cursor, Claude Code, and Claude Desktop; API tests generated from OpenAPI/Swagger; SmartBot Web from scratch.
Same ANF analyzer JSON object in the middle (May 2026). Switch client to inspect how Slack, SmartBot Web, and IDE agents render the same check.
Shared structured payload
{
"analyzer": "anf_region_health",
"status": "failed",
"failed_checks": ["quota_headroom"],
"skipped_locations": ["westus2"],
"empty_fields": ["notes"],
"dry_run": true
}
Client output
anf_region_health · failed
Failed check: quota_headroom
Skipped location: westus2 · Empty: notes
Cleanup path: dry_run: true (0 deleted)
Slack threads quote the exact same check fields so engineers in every time zone read one answer.
| Field | Value |
|---|---|
status | failed |
failed_checks | quota_headroom |
skipped_locations | westus2 |
empty_fields | notes |
dry_run | true (0 deleted) |
SmartBot Web renders the same analyzer state, cleanup dry-run, and analytics without a private copy.
Tool: anf_region_health
First reply: quotes the raw JSON result verbatim - failed, westus2 skipped, dry_run: true.
Guardrail: mutating tools run dry-run first and stop until a person confirms delete.
An agent does not paraphrase a failed check into "mostly fine" or a dry-run into "cleaned up."
Viti is Icelandic for lighthouse. It is the new assistant on marioruiz.net: it calls my public MCP servers, answers in Spanish, English and Icelandic, and can listen and speak.
An assistant that only remembers the chat will guess. Two public MCP servers on this site connect to Claude, ChatGPT, Cursor or VS Code. A server it can call has to show its work.
When one assistant becomes several, the hard part is the roster: a role, a trial, one owner per task, access that ends with the role, and a record of why one was retired.
AI agents and MCP on top of senior backend scope: I own SmartBot for 800+ employees across 30+ time zones and ship MCP + JSON API surfaces, LLM sessions in Slack, and API tests from OpenAPI. I designed and built SmartBot, and received the 2021 NetApp Public Cloud Excellence Award (Most Innovative Employee).
The same ANF analyzers Slack already uses, as JSON API and MCP. One payload, three fronts: Slack, SmartBot Web, and IDE agents (Cursor, Claude Code, Claude Desktop).
Team ChatGPT sessions with a team knowledge base as context; natural language for 50+ commands (ANF Sessions). 40k+ LLM session calls.
Slack + web platform used daily across 150+ Slack channels. 250+ commands, 800+ employees, 30+ time zones covering team management and product commands, cutting repetitive ops work.
A SmartBot cleanup service people run in Slack - plus web UI, REST and dry-run - to delete stale Azure test resources. It contributed to $2M+ in cloud cost reduction.
Tools generating API tests from product OpenAPI/Swagger (TypeSpec-aligned, multi-region) so the suite stays in step with the spec; SmartBot Web from scratch for tests, cleanup, analytics, and tutorials.
Same command and API-test pattern on Azure NetApp Files and Cloud Volumes Service (AWS, Google Cloud), plus scheduled routines for region health, test-location analysis, and workflow triggers.
October 2007 - August 2018 · Iceland. Company-wide test framework and internal tools (Ruby, Sinatra): browsers, REST, databases, queues. Several public libraries started here, and the HipChat tool that became Slack SmartBot.
Libraries on RubyGems. Combined downloads 3.5M+ as of Sep 2026.
Test-data factories from patterns / regex
Extensible Slack automation engine (core of SmartBot)
HTTP/REST client for testing
Hash/JSON generation for API tests
Swagger / OpenAPI to request hashes (3.1)
Generate RSpec tests from those request hashes
Elapsed-time measurement for slow spots in code
OpenAPI 3 parser (maintained fork)
Three interactive pieces from Experiments you can run directly in the browser.
Since childhood I have written poems, made music, played the trumpet and cooked. It is the same habit as building tools. In 2026 I published a bilingual book of poems and music albums under a pen name. The music has passed 600k plays since July 2026.
At ten I networked 8-bit MSX computers over sound: they negotiated who got the printer and sent the job - all in audio.