Back to Details

firecrawl-developer-search

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

firecrawl developer search

Firecrawl 开发者搜索

Via the Firecrawl MCP (preferred)

通过 Firecrawl MCP（首选）

This plugin bundles the Firecrawl MCP server. When it's connected, prefer the native 
firecrawl_developer_search
 tool over the CLI — it needs no local install and authenticates through the plugin's one-time browser sign-in. The 
firecrawl
CLI commands below are the fallback for when the MCP isn't connected.
Semantic search over indexed developer sources. Each result carries an 
id
, a 
url
, a title, and the matched passages in markdown, so tables and code blocks survive intact.
The 
id
prefix tells you what kind of source a hit is: 
doc:
, 
issue:
, 
pull_request:
, or 
readme:
.
此插件内置 Firecrawl MCP server。连接后，应优先使用原生的 
firecrawl_developer_search
 工具，而不是 CLI——它无需本地安装，并通过插件的一次性浏览器登录完成身份验证。以下 
firecrawl
CLI 命令用于 MCP 未连接时的回退方案。
对已索引的开发者资料源进行语义搜索。每个结果都包含一个 
id
、一个 
url
、一个标题以及 Markdown 格式的匹配段落，因此表格和代码块能够完整保留。
id
前缀表示命中结果所属的资料源类型：
doc:
、
issue:
、
pull_request:
或 
readme:
。

When to use

何时使用

  • The question is about code behaviour, a library or framework, an API contract, an error message, or a known bug
  • You want the issue that reported it, the merged PR that fixed it, or the doc page that defines the contract — not a blog post describing it
  • Before falling back to a general web search: reach for the index first, the open web second
Not for this skill: comparisons and opinions ("X vs Y"), news and outages, or a project with no public repo or indexed docs. Those are firecrawl-search plus firecrawl-scrape questions.
  • 问题涉及代码行为、库或框架、API 约定、错误信息或已知 bug
  • 你需要查找报告问题的 issue、修复问题的已合并 PR，或定义 API 约定的文档页面，而不是介绍该问题的博客文章
  • 在回退到通用 Web 搜索之前，应先查询索引，再搜索开放 Web
此 skill 不适用于：比较和观点（“X 与 Y 对比”）、新闻和服务中断，或没有公开仓库及已索引文档的项目。此类问题应使用 firecrawl-search 和 firecrawl-scrape。

Quick start

快速开始

MCP (preferred):
text
firecrawl_developer_search(query: "tokio select cancellation safety", k: 10)
firecrawl_developer_search(query: "agent skill frontmatter conventions", skills: "only")
CLI fallback:
bash
undefined
MCP（首选）：
text
firecrawl_developer_search(query: "tokio select cancellation safety", k: 10)
firecrawl_developer_search(query: "agent skill frontmatter conventions", skills: "only")
CLI 回退方案：
bash
undefined

Ask a developer question

提出一个开发者问题

firecrawl developer "axum middleware ordering" --limit 10 -o .firecrawl/dev-axum-middleware.json --json
firecrawl developer "axum middleware ordering" --limit 10 -o .firecrawl/dev-axum-middleware.json --json

Search a literal error string plus the library name

搜索错误字符串及对应的库名称

firecrawl developer "ERR_REQUIRE_ESM in vitest config" -o .firecrawl/dev-esm-error.json --json

Put all scoping intent in the query text — the repo slug, the library name, the version, the error string. Semantic retrieval handles the scoping; there are no repo or type flags on the MCP tool or the CLI.
firecrawl developer "ERR_REQUIRE_ESM in vitest config" -o .firecrawl/dev-esm-error.json --json

将所有限定范围的信息放入查询文本中——例如仓库 slug、库名称、版本和错误字符串。语义检索会处理范围限定；MCP 工具和 CLI 均不提供仓库或类型筛选标志。

Options

选项

Option (MCP)Option (CLI)Description
query
<query>
Natural-language question or search phrase (required)
k
--limit <n>
Number of ranked results, 1–100 (default 10)
skills
—Set to 
"only"
to search only agent-skill files
—
-o, --output <path>
Output file path
—
--json
/ 
--pretty
JSON output, compact or pretty
选项（MCP）选项（CLI）说明
query
<query>
自然语言问题或搜索短语（必填）
k
--limit <n>
排序结果数量，1–100（默认 10）
skills
—设为 
"only"
可仅搜索 Agent skill 文件
—
-o, --output <path>
输出文件路径
—
--json
/ 
--pretty
JSON 输出，紧凑或格式化显示

Tips

提示

  • Match the query to the question. A literal error message wants the error string plus the library name. A conceptual "how do I do X" wants the whole question in natural language.
  • Volatile parts hurt a match. If an error string returns nothing, strip paths, line numbers, ids, and addresses and retry — the invariant middle of the message is what's indexed.
  • A merge supersedes a report. When an issue and a pull request disagree, the merged PR is the current behaviour. Read the resolution before answering, never the opening report alone.
  • Quote the passage, cite the 
    url
    .     The passages are the evidence — hand them over rather than paraphrasing into a claim the user can't check. 
    title
    is often absent on 
    doc:
    hits; fall back to 
    url
    .
  • Raise 
    k
    before giving up, and scrape when a hit is the right page.     If a result is clearly the right page but the passage is the wrong part of it, scrape its 
    url
    for the whole thing.
  • Already running a web search? 
    firecrawl_search(query, categories: ["developer"])
    folds developer hits into the same call as web results (they come back tagged 
    category: "developer"
    ). It exposes no passage control, so prefer the dedicated tool when the question is purely a code question.
  • With the CLI, write results to 
    .firecrawl/
    with 
    -o
    to keep the context window clean. Naming convention: 
    .firecrawl/dev-{topic}.json
    .
  • 让查询与问题相匹配。 搜索字面错误信息时，应同时提供错误字符串和库名称。搜索概念性的“如何完成 X”时，应使用完整的自然语言问题。
  • 易变内容会降低匹配效果。 如果错误字符串没有返回结果，请移除路径、行号、id 和地址后重试——索引收录的是错误信息中保持不变的核心部分。
  • 已合并的 PR 比问题报告更具权威性。 当 issue 与 pull request 的信息不一致时，以已合并的 PR 所体现的当前行为为准。回答前应阅读问题解决方案，不能只依据最初的报告。
  • 引用段落，并标注 
    url
    。     段落才是证据——应将其直接呈现给用户，而不是改写成用户无法核实的结论。
    doc:
    命中结果通常没有 
    title
    ，此时可回退使用 
    url
    。
  • 不要过早放弃，应先提高 
    k
    ；命中页面正确时再进行抓取。     如果某个结果明显对应正确页面，但匹配段落不是所需部分，请使用 scrape 抓取其 
    url
    以获取完整内容。
  • 如果已经在执行 Web 搜索，
    firecrawl_search(query, categories: ["developer"])
    会将开发者资料源的结果合并到同一次 Web 搜索调用中，并以 
    category: "developer"
    标记。该方式无法控制返回的段落，因此对于纯代码问题，应优先使用专用工具。
  • 使用 CLI 时，可通过 
    -o
    将结果写入 
    .firecrawl/
    ，以保持上下文窗口整洁。命名约定为 
    .firecrawl/dev-{topic}.json
    。

See also

另请参阅

  • firecrawl-search — general web search, and the 
    categories: ["developer"]
    blend
  • firecrawl-scrape — read a full page behind a promising hit
  • firecrawl-agent — autonomous multi-source research when one search won't settle it
  • firecrawl-search——通用 Web 搜索，以及 
    categories: ["developer"]
    混合查询
  • firecrawl-scrape——读取 promising hit 背后的完整页面
  • firecrawl-agent——当一次搜索无法确定答案时，执行自主的多资料源研究