firecrawl-plugin

Firecrawl plugin for Hermes Agent

Generated from Firecrawl's plugin source. Do not edit this repository by hand; changes made here are overwritten by the next sync.

Firecrawl is the context API for AI agents. This plugin connects Hermes Agent to the hosted Firecrawl MCP server and adds Firecrawl skills, so Hermes can search the web, read pages, collect structured data, interact with live pages, crawl sites, run research and track changes. It is an Agent Plugins 1.0.0 package: a manifest, an MCP connection and Markdown skills, with no local code or runtime dependencies.

Hermes already ships Firecrawl as a backend for its built-in web_search and web_extract tools. This plugin is separate from that backend and does not replace it: it adds the full Firecrawl tool set (crawl, map, interact, agent, monitors, parse, Alexandria data providers, people and company search, and developer, paper and US government search) and skills that route requests to the right tool. Both can use the same FIRECRAWL_API_KEY.

What's included

mcp.json: the hosted Firecrawl MCP server at https://mcp.firecrawl.dev/v2/mcp-oauth, as the firecrawl server
skills/get-started: checks the connection, then gets you to a useful first result
skills/firecrawl: routes each request to the right Firecrawl tool, with a reference per operation
skills/firecrawl-deep-research: cited, multi-source investigations, comparisons, prospect discovery and enrichment
Workflow skills: brand-identity, business-contact-research, design, enrich-lead, fantasy-sports, find-leads, flights-and-hotels, job-postings, local-businesses, markets-and-filings, news-coverage, prices-and-plans, property-listings, read-docs, seo-audit and video-transcripts

Requires Hermes Agent 0.22.2 or later, with its MCP support installed.

Install

hermes plugins install firecrawl-plugin

Portable packages are installed disabled. Enable it in the profile you chat from:

hermes plugins enable firecrawl-plugin

To install straight from this repository instead of the catalog:

hermes plugins install firecrawl/firecrawl-hermes-plugin --no-enable

Connect your Firecrawl account

Both methods use the same endpoint, https://mcp.firecrawl.dev/v2/mcp-oauth. Choose one, and use the same Hermes profile for installation, configuration, sign-in and chat.

Browser sign-in (OAuth, recommended)

Hermes runs OAuth for servers declared in your profile's config.yaml with auth: oauth, and the portable plugin format has no field for it, so browser sign-in takes one native entry. It takes precedence over the plugin's server of the same name. Replace any existing mcp_servers.firecrawl entry as a whole and keep your other servers:

mcp_servers:
  firecrawl:
    url: https://mcp.firecrawl.dev/v2/mcp-oauth
    auth: oauth
    trust: untrusted
    strict_redirect_headers: true
    headers:
      X-Firecrawl-Origin: firecrawl-hermes-plugin

Keep trust: untrusted: this entry replaces the plugin's server, trust setting included (see Approvals).

Then sign in from a terminal:

hermes mcp login firecrawl

Hermes stores and refreshes the tokens; no FIRECRAWL_API_KEY is needed. Do not add an Authorization header to this entry. Cron jobs and other unattended runs cannot complete a browser sign-in, so sign in first, or use an API key for unattended work.

API key

Create a key at firecrawl.dev/app/api-keys and save it to the active profile:

hermes config set FIRECRAWL_API_KEY fc-your-key

With no native entry, the plugin's mcp.json sends the key as a bearer token. If you already use Firecrawl as Hermes's web backend, the same key works. Keep the key out of chat, URLs and this repository.

Use

Ask for what you need in plain language, or start with the get-started skill. Hermes names the tools mcp__firecrawl__<tool>, for example mcp__firecrawl__firecrawl_search and mcp__firecrawl__firecrawl_scrape; the skills refer to them by the tool name alone (firecrawl_search), and Hermes loads them as firecrawl-plugin:<skill>.

"Search for this week's announcements about Agent Plugins and summarize them with links."
"Scrape https://docs.firecrawl.dev and list the SDKs it documents."
"Map docs.example.com and crawl the /api section."
"Find 20 Series A fintech companies in London with their websites."
"Use firecrawl-deep-research to compare the three largest vector databases on pricing and limits."

Approvals

The plugin marks the firecrawl server trust: untrusted, so Hermes asks before it runs any tool the server does not annotate as read-only: for example firecrawl_crawl, firecrawl_interact (live browser actions, including form submissions) and creating, updating, running or deleting monitors (recurring jobs that spend credits). Search, scrape, map, parse, research with firecrawl_agent, people and company lookups and the other read-only tools run without a prompt. Approve a tool once, for the session, or always.

Capabilities

Search: web, news and image search with query-relevant excerpts, plus matching Alexandria data providers
Scrape: any URL to clean Markdown, JSON with a schema, a grounded answer or verbatim highlights
Alexandria: discover catalogued data providers and execute them for structured records
People and companies: search by title, employer, location or industry, and enrich known people or companies
Agent: autonomous multi-source collection into structured data
Map and crawl: list a site's URLs or collect a bounded section
Interact: clicks, forms and navigation in a live browser session
Parse: PDFs and documents, including local files through a hosted upload
Monitor: scheduled change detection for pages and web searches
Developer, paper and government search: repositories and docs, research papers with full text, and US law and regulation

What this plugin does on your machine and the network

Sends the queries, URLs, prompts and files you ask Firecrawl to process to https://mcp.firecrawl.dev over HTTPS, authenticated with your OAuth sign-in or API key. Tool calls use your Firecrawl account's credits; firecrawl_credit_usage reports usage.
Sends X-Firecrawl-Origin: firecrawl-hermes-plugin so Firecrawl can attribute requests to this plugin. There is no separate telemetry.
firecrawl_parse uploads a local file only when you ask it to parse that file.
Asks Hermes to gate the server (trust: untrusted), so tools that change state or start credit-spending jobs need your approval first.
Runs no local code, shell commands, background processes or self-updater, and reads no credential files. Hermes owns the MCP connection, credentials and approvals.
The hosted server's tools can change independently of this package; the live tool schema is the authority.

Support

Docs: docs.firecrawl.dev
Issues: github.com/firecrawl/firecrawl-mcp-server/issues
Email: help@firecrawl.dev

License

MIT

README from the reviewed commit 561fcec; it updates when the author re-pins.

About