Firecrawl MCP Setup: Start Keyless, Then Add OAuth or an API Key
Start with Firecrawl's current keyless remote MCP endpoint: it lets you try Search, Scrape, and Parse within daily limits without an account or API key. Move to browser OAuth for team-plan access, or to the API-key endpoint when you need more tools, higher volume, or noninteractive execution.
Start with Firecrawl's current keyless remote MCP endpoint: it lets you try Search, Scrape, and Parse within daily limits without an account or API key. Move to browser OAuth for team-plan access, or to the API-key endpoint when you need more tools, higher volume, or noninteractive execution.
Table of contents
- Which Firecrawl MCP connection should you choose?
- How do you connect without an account or key?
- When should you switch to OAuth or an API key?
- How do you verify search, scrape, and parse safely?
- How do you control crawling scope, data, and tool count?
- How do you fix quotas, blocked sites, timeouts, and missing tools?
- FAQ
Which Firecrawl MCP connection should you choose?
Start with keyless remote access for a quick workflow check, then choose OAuth, an API-key endpoint, or the local package based on how your client and workload operate. Firecrawl documents keyless, browser OAuth, API-key remote, and local npx paths in its MCP server setup guide.
| Need | Recommended connection | Why |
|---|---|---|
| Try web search and extraction | Keyless remote | No account or key; includes Search, Scrape, and Parse within daily limits |
| Use team-plan access | Browser OAuth | Connects through an interactive browser authorization flow |
| Run unattended workflows | API-key remote | Uses the documented API-key endpoint for noninteractive execution |
| Keep the server local | firecrawl-mcp through npx | The official package supports local operation and configuration |
The keyless path is the best starting point when you are deciding whether Firecrawl MCP fits a daily research or documentation workflow. For a practical directory entry and related configuration options, see Firecrawl MCP on MCPtrove.
MCP separates the host or client from the server that provides tools. That makes the connection choice important: the client determines how you add the server, while Firecrawl determines which tools and access mode are available. The MCP architecture documentation explains this relationship.
How do you connect without an account or key?
Choose Firecrawl's keyless remote option in an MCP client that supports remote servers, then approve the connection without entering an account or API key. The keyless route provides Search, Scrape, and Parse within daily limits.
Use this sequence:
- Open the MCP server configuration area in your client.
- Select the remote-server option and follow Firecrawl's keyless MCP instructions.
- Confirm that the Firecrawl server appears as connected.
- Ask the client to perform a narrow search.
- Scrape one result URL, then parse a small, explicit extraction task.
Keep the first request small. A single search query and one target page tell you whether the connection works without introducing unnecessary crawling scope. If the client asks for an OAuth login or an API key during this path, check that you selected the keyless route rather than the OAuth or API-key endpoint.
The keyless route is intended for proving the basic workflow, not for assuming unlimited access. Daily limits apply, and the available tool set is narrower than a configured account-based connection. Record which of Search, Scrape, and Parse are visible after setup so missing tools are easy to distinguish from a failed connection.
When should you switch to OAuth or an API key?
Switch to browser OAuth when you need team-plan access; use the API-key remote endpoint when you need more tools, greater volume, or noninteractive execution. Firecrawl documents the OAuth endpoint as https://mcp.firecrawl.dev/v2/mcp-oauth and the API-key endpoint as https://mcp.firecrawl.dev/v2/mcp.
Choose browser OAuth when a person can complete an interactive authorization step and the workflow should use the access available through a team plan. The OAuth endpoint is:
https://mcp.firecrawl.dev/v2/mcp-oauth
Choose the API-key endpoint when the MCP client or automation process must connect without a browser step. The API-key endpoint is:
https://mcp.firecrawl.dev/v2/mcp
The decision is operational:
- Keyless is for a quick start with no account or key.
- OAuth is for an interactive team-plan connection.
- API-key remote is for unattended execution and workloads that need more access.
- Local
firecrawl-mcpis for a locally managed server process.
Do not switch merely because the first request is unfamiliar. First confirm that the server is connected, the selected endpoint is correct, and the requested operation is one of the tools available to that connection. Firecrawl's MCP repository is the appropriate reference for the official local server package.
How do you verify search, scrape, and parse safely?
Verify the connection with one narrow search, one known result, and one limited parse request. This staged sequence keeps the requested data and crawling scope clear while showing which operation is failing.
Start with Search:
- Use a focused query with a clear topic.
- Check that the response contains usable result links.
- Select one result rather than sending a broad list for extraction.
Continue with Scrape:
- Submit the selected URL.
- Confirm that the response corresponds to the requested page.
- Check the page title, main content, and source domain before using the result.
Finish with Parse:
- Ask for a small set of fields or a concise structure.
- Compare the parsed output with the scraped content.
- Stop if the requested fields are absent or the page does not match the intended source.
This sequence also helps isolate failures. A successful Search followed by a failed Scrape suggests a target-page or site-access issue. A successful Scrape followed by an incomplete Parse suggests that the extraction request is too broad or that the requested fields are not present.
Keep authorization and data handling in scope throughout the test. The MCP security best practices document provides the relevant guidance for treating tool connections, inputs, and returned data carefully.
How do you control crawling scope, data, and tool count?
Control scope by starting with one query and one URL, limit the requested fields, and expose only the tools your workflow needs. The local firecrawl-mcp package supports tool filtering as well as retry and rate-limit configuration.
A practical scope policy looks like this:
- Search only for the topic required by the task.
- Scrape only selected URLs.
- Parse only named fields or sections.
- Avoid sending a whole result set when one source answers the question.
- Keep the first request small enough to inspect manually.
- Add tools only when the workflow has a defined need for them.
Tool filtering matters when a local server exposes more capability than a client or user needs. Fewer visible tools can make the available workflow easier to understand and reduce accidental requests. The official package documentation and repository provide the implementation reference for local setup.
For search-focused workflows, the MCPtrove web-search capability page is a useful next step for comparing related tools and use cases. For documentation work, see MCP servers for documentation to connect the Firecrawl workflow to a more specific content task.
Use the same principle for returned data: request only what the next step needs. A short, structured parse is easier to inspect than a large page response, especially when you are testing a new connection or working within daily limits.
How do you fix quotas, blocked sites, timeouts, and missing tools?
Fix these issues by identifying whether the problem is a daily limit, a target-site response, a request scope, or the selected connection's tool set. Then reduce the request, check the documented limits, or switch connection types only when the workload requires it.
| Symptom | Likely cause | Next step |
|---|---|---|
| Requests stop after repeated use | Daily or rate limit | Review Firecrawl's rate-limit documentation, reduce request volume, and retry within the documented limits |
| One domain does not return usable content | Target-site response or access issue | Test one different permitted URL and keep the scope narrow |
| A request times out | Task is too broad | Try one URL, a smaller extraction, or a separate Search, Scrape, and Parse sequence |
| Search works but another tool is absent | Keyless tool set is limited | Use OAuth, the API-key endpoint, or the local package when the workflow needs more tools |
| Local setup shows too many tools | Tool configuration is broad | Apply the official package's tool-filtering configuration |
| Unattended execution cannot complete | Browser authorization is required | Use the API-key endpoint instead of an interactive OAuth connection |
Avoid solving a failure by disabling authentication, TLS, or permission checks. A smaller request and a correctly selected connection provide a clearer diagnosis.
If configuration remains unclear, use MCPtrove's Config Doctor as the practical next step for checking the server setup. Keep the original endpoint, connection mode, and failing operation visible while troubleshooting so the issue can be reproduced.