Best MCP Servers for Hermes Agent in 2026

Hermes Agent runs as your own always-on process rather than a SaaS subscription, and it draws a real line between "web search backend" and "MCP tool server" — two config sections, two different vendor lists. Nobody's laid out which of the major web-data vendors sits in which list. This does, from Hermes's own docs and each vendor's own integration guide.

Hermes Agent's two paths to web data

Native web-search backends. Hermes ships eight pluggable web-search/extract backends configured in ~/.hermes/config.yaml's web: block — Firecrawl (the default), SearXNG, Brave, DuckDuckGo, Exa, Parallel, and xAI. No MCP protocol involved; the agent's built-in web_search and web_extract tools call straight into whichever backend is configured.

A native MCP client. Separately, Hermes's tool-server layer speaks MCP directly — stdio and HTTP transports, OAuth flows, per-tool security rules, auto-reload when config changes, and it can even run as an MCP server itself so other MCP clients (Claude Code, Cursor, Codex) can call Hermes's own messaging tools. Servers go under mcp_servers: in config.yaml:

mcp_servers:
  # stdio (local process)
  bright-data-web-mcp:
    command: "npx"
    args: ["-y", "@brightdata/mcp"]
    env:
      API_TOKEN: "your_bright_data_key"
      PRO_MODE: "true"
  # HTTP (remote)
  crawlora:
    url: "https://mcp.crawlora.net/mcp"
    headers:
      Authorization: "Bearer your_crawlora_key"

Three ways to add a server: hand-edit config.yaml, hermes mcp for an interactive picker over a curated catalog (hermes mcp install <name>), or hermes mcp add <name> for a custom one. The web dashboard's MCP page offers the same thing through an ADD SERVER modal if you'd rather not touch YAML.

What to evaluate, for Hermes Agent specifically

• Backend or tool server: if the job is web_search / web_extract, check the native backend list first — it's less config and no local process for options like Firecrawl. If the job needs a named platform's structured data, that's the mcp_servers: layer.
• Self-hosted, always-on context: Hermes runs as your own persistent process, so a stdio MCP server is a long-running child process on whatever host runs Hermes — an HTTP server avoids that.
• Parallel tool execution: Hermes runs MCP tools sequentially by default; if a server's tools are safe to run concurrently (read-only, independent), you can opt into parallel execution per server.
• Bidirectional MCP: Hermes can also expose its own messaging tools as an MCP server to other agents — worth knowing if your stack already runs Claude Code or Codex alongside Hermes.
• Self-hosted target for Firecrawl specifically: FIRECRAWL_API_URL lets the native backend point at a self-hosted Firecrawl instance instead of the hosted API.

The contenders

Firecrawl: already the default, no MCP required

Firecrawl is Hermes's default web backend — set FIRECRAWL_API_KEY in ~/.hermes/.env and Hermes auto-detects it, no config.yaml edit needed at all. The hermes tools interactive wizard writes the web: block for you if you'd rather go through a menu. A June 2026 change removed a redundant LLM-summarization step from the pipeline, so clean Firecrawl markdown now passes straight to the agent. Choose it as the baseline — it's already there unless you've switched backends.

Bright Data: a real MCP server, 69 tools

Bright Data connects through Hermes's actual MCP layer, not the native backend list. Through the dashboard: open the MCP page, ADD SERVER, transport stdio, command npx, args @brightdata/mcp, then API_TOKEN (required) and PRO_MODE=true (optional) as environment variables. The equivalent config.yaml entry is shown above. Pro mode currently exposes the full 69-tool set across named e-commerce, social, business, and finance platforms plus proxy and browser infrastructure; the free Rapid tier gives a couple of tools and 5,000 requests/month. Choose it for stubborn, anti-bot-protected targets, and accept a second long-running process next to Hermes's own.

Zyte: no Hermes Agent presence yet

Zyte's Claude Code plugin, GitHub Copilot CLI plugin, and Codex add-on don't extend to Hermes Agent — nothing turned up in Hermes's integrations docs, Zyte's own release notes, or either vendor's blog. If Hermes is your framework, Zyte isn't currently in the running.

Crawlora: HTTP transport, no process, 236 platform groups

Crawlora's MCP server is the HTTP-transport entry in this comparison — a url and a headers.Authorization bearer token, no npx process for Hermes to keep alive alongside its own always-on loop:

mcp_servers:
  crawlora:
    url: "https://mcp.crawlora.net/mcp"
    headers:
      Authorization: "Bearer your_crawlora_key"

The current catalog covers 2,021 MCP tools across 236 platform groups — search, maps, e-commerce, social, finance, travel, jobs, media, app stores, and reviews — returning normalized JSON, billed only on success, 2,000 free credits a month, no card. Choose it when Hermes needs a number or record from a named platform rather than a page to read, and you'd rather not add another local process to a host already running the agent itself.

How to choose, by job

• Improving Hermes's default web search with the least setup → Firecrawl, already the default backend.
• Extracting from a stubborn, named site at real volume → Bright Data's MCP server, one more process to run.
• Pulling a price, ranking, listing, or filing from a known platform without adding another process to this host → Crawlora MCP.
• Owning a Scrapy codebase instead of calling tools live → Claude Code, Copilot, or Codex for now, not Hermes.
• Running Hermes as a persistent self-hosted process → prefer HTTP-transport MCP servers over stdio ones where you can, to avoid stacking child processes on the same host.

Add structured web data to Hermes Agent

2,021 MCP tools across 236 platform groups, one API key, billed only on success. 2,000 free credits a month, no card.

Set up Crawlora MCPBrowse the API catalog

Sources

Related reading

• Best MCP Servers for OpenClaw in 2026: another self-hosted agent framework, same vendors, a different native-vs-MCP split.
• Best MCP Servers for OpenAI Codex in 2026.
• Best Web Scraping MCP Servers & Skills for Claude Code in 2026.
• Crawlora vs Firecrawl · Crawlora vs Bright Data · Crawlora vs Zyte.
• Give your AI agent live web data with MCP.