Firecrawl MCP server in Continue.dev web data in your open-source assistant.
Continue configures MCP servers in YAML — either globally in config.yaml or as per-project files under .continue/mcpServers/. Here's the exact block for Firecrawl and the YAML-specific failure modes.
Continue.dev is the open-source assistant that runs in VS Code and JetBrains, and since it speaks MCP, Firecrawl MCP Server (github.com/firecrawl/firecrawl-mcp-server, 6.5k stars) plugs straight in — giving your agent web search, page scraping, full-site crawling, and structured extraction.
Unlike most MCP clients, Continue's configuration is YAML, not JSON. That changes the failure modes: indentation matters, and the block goes in ~/.continue/config.yaml or in a standalone file under .continue/mcpServers/ in your project.
Prerequisites
Install Node.js 18+ (Continue spawns the server with npx) and grab a Firecrawl API key from firecrawl.dev — keys start with fc-. One Continue-specific requirement worth knowing before you debug anything: MCP tools are only available in agent mode, not in plain chat, so make sure your selected model supports tool use.
Add the mcpServers block
For a global setup, open ~/.continue/config.yaml and add the block below. For a per-project setup, drop the same server definition into a YAML file at .continue/mcpServers/firecrawl.yaml instead — useful when only one repo needs scraping.
mcpServers:
- name: Firecrawl
command: npx
args:
- "-y"
- firecrawl-mcp
env:
FIRECRAWL_API_KEY: fc-YOUR-API-KEYReload and test
Reload the Continue extension (Developer: Reload Window in VS Code is the sure way), switch the input to agent mode, and check the tools indicator — Firecrawl's scrape, search, crawl, and extract tools should be listed. A quick test prompt: "search the web for the latest Node LTS version and cite the page you scraped."
Each tool has a distinct job, and Continue's agent picks between them: scrape fetches one URL as clean markdown, search runs a web query and can scrape the results, map lists a site's URLs without fetching content, and crawl walks a whole site — the expensive one, since every page it touches consumes Firecrawl credits.
Want to confirm what tools this server should expose before you debug? Its live Loomal listing at https://loomal.ai/marketplace/firecrawl-mcp-server shows the probed tool list.
Troubleshooting in Continue.dev
Tools missing: confirm you're in agent mode first — MCP servers are invisible in normal chat, and that explains the majority of "it doesn't work" reports. Second, YAML: mcpServers is a list, so each server starts with a dash; two-space indentation throughout; and quote the "-y" argument so YAML doesn't parse it as something else.
Server fails to start: if you're still on Continue's older config.json format, the YAML block won't be read — migrate to config.yaml or use the equivalent JSON structure for your version. PATH issues show up as spawn errors; launch your editor from a terminal so npx resolves, or use its absolute path. And if tools run but return auth errors, the fc- key in env is wrong or expired.
FAQ
How do I install Firecrawl in Continue.dev?
Add an mcpServers list entry to ~/.continue/config.yaml (or a YAML file in .continue/mcpServers/) with command npx, args -y and firecrawl-mcp, and FIRECRAWL_API_KEY in env. Reload the Continue extension and switch to agent mode to see the tools.
Where does Continue store MCP configuration?
Globally in ~/.continue/config.yaml, or per project in .continue/mcpServers/*.yaml files. The per-project form is handy for keeping a scraping server scoped to the one repo that needs it.
Why don't Firecrawl's tools show up in chat?
Continue only exposes MCP tools in agent mode. Plain chat ignores them entirely, and so do models that don't support tool calling. Switch the mode selector to agent and pick a tool-capable model before assuming the config is broken.
My YAML looks right but the server never starts — what else can it be?
Check that you're on the config.yaml format (older Continue versions used config.json and won't read YAML), verify npx is on the PATH the editor inherits, and watch for tab characters — YAML requires spaces, and a single tab silently invalidates the file.
More MCP servers for Continue.dev.
Browse scraping, search, and dev-tool servers with live tool lists.