# Your app via MCP (community wiki)

Zdroj: https://hermes-ai.cz/en/wiki/integrations/your-app-via-mcp/
Ověřeno proti oficiální dokumentaci: 17. 10. 2026
Oficiální stránka: https://hermes-agent.nousresearch.com/docs/guides/use-mcp-with-hermes
Language: English. AI version — paste the whole text into your agent.

## When MCP
Yes: an existing MCP tool, a clean layer onto a company system/DB, fine control over visible tools.
No: a built-in tool already does it; the server exposes a huge dangerous surface; one narrow thing where
a small script is simpler.

## Configuration (~/.hermes/config.yaml)
mcp_servers:
  name:
    command/args/env        # stdio
    url/headers             # HTTP (Authorization: "Bearer ${TOKEN}")
    ssl_verify: true|false|CA path
    enabled, timeout, connect_timeout, keepalive_interval, lazy, transport: sse
    tools: {include: [...], exclude: [...], resources: true, prompts: true}

## Procedure
1) run the server standalone · 2) config.yaml · 3) hermes mcp list · 4) try one tool
· 5) only then cron/community. If the server fails alone, inside Hermes it looks like an agent bug.

## Watch out
Many tools = a slower, pricier agent -> tools.include/exclude · secrets via ${VAR}, value in .env/vault
· timeout/keepalive when calls drop · dangerous tools in exclude · lazy: true for a fast start.

## Our practice
App database: reads yes, writes with confirmation. Sentry: the correct regionUrl (de.sentry.io).
Library docs: per library only, never "everything about everything".