DEV Community

Renato Marinho
Renato Marinho

Posted on

Vinkius for Agents: The MCP Catalog That Stops Your Bots Guessing

#ai #devrel #mcp #tutorial

Vinkius for Agents: The MCP Catalog That Stops Your Bots Guessing

The Model Context Protocol just quietly became the USB-C of AI. But here's the friction nobody's talking about: discovering the right tool in the first place. We built Vinkius to kill that friction.

The problem we hit on a Friday night

You know the moment. You're building an agent. It needs to do one specific thing — say, validate a voice channel or parse a Dev.to engagement report. You go to the MCP server list on GitHub, and you're staring at 9,000+ entries. No filtering. No semantics. Just a flat list of slugs.

You start grep-ing. You misread a schema. You wire it up, and the agent hallucinates because it's calling the wrong tool. Now your agent is confidently wrong, and your users are watching.

That's the workflow we wanted to eliminate.

What Vinkius actually does

Vinkius isn't another MCP directory. It's a semantic index over the MCP ecosystem. Each server you plug in comes with its real tools, input schemas, reliability records, and a declared contract — so when your agent (or your DevRel pipeline) hits the catalog, it gets typed tools back, not a link to a README.

The killer detail: the catalog is itself an MCP endpoint. Your agent doesn't need to scrape a website. It calls:

from vinkius_catalog import search_catalog

result = search_catalog(query="validate DevRel voice",
                        page=1)
# result -> [{slug, title, short_description, tools, ...}]
Enter fullscreen mode Exit fullscreen mode

One call. Typed schema. Zero markdown parsing. This is the workflow change that matters.

Why developers should care

Pain Old workaround Vinkius
Find the right MCP for a task Browse GitHub, read docs, guess search_catalog with a natural-language query
Know a server's reliability Trust the README reliability= record on each connector
Wire it into an agent Manually parse a markdown file Direct invoke_tool against a concrete server_id
Validate your own content (DevRel voice, API contracts) Manual review Call devrel-voice-prover or api-design-prover in the same run

The last row is the one that sold me. We publish DevRel content and tutorials on Dev.to and Hashnode. Until now, "does this sound like a senior engineer or a press release" was a vibes check. Now it's a tool call:

vinkius_catalog.invoke_tool(
  server_id="019e58c7-4612-7293-bb57-08d6780c8c2b",  # devrel-voice-prover
  tool="validate_devrel_voice",
  arguments={"action":"default", "contentType":"...", "communityContext":"..."}
)
Enter fullscreen mode Exit fullscreen mode

If it fails, you fix it before you publish. That's the workflow change.

Get started in two minutes

  1. Grab the endpoint. Vinkius is an MCP server — you connect your agent to it the same way you connect to any other.
  2. Search. One query, get typed tools. Install what matches.
  3. Invoke. Same run, same contract. No scraping, no guessing.

The whole thing is open, and the catalog is public. Start with the two tools we ship out of the box that every developer should have wired in:

  • vinkius-mcp-catalog — discovery + typed invocation
  • devrel-voice-prover — stop your AI from writing press releases to a developer audience

If you're building agents and you've ever wasted an afternoon reading 40 MCP READMEs looking for one tool, that afternoon is the reason Vinkius exists. The catalog's live, the endpoint is public, and the contract is typed. Wire it in.

How this was produced

This article was written against the devrel-voice-prover contract and validated by the tool before publication — the same workflow it describes. It's the proof-of-concept: the tool that validates the article is the tool the article recommends.

About Vinkius: a semantic index and typed connector layer over the Model Context Protocol ecosystem. Open catalog, live endpoint, public contract.

Top comments (0)