Firemark

com.tryfiremark由 tryfiremark.com 发布官方发布者v1.6.0更新于 Nov 7, 2026

Read how often ChatGPT, Perplexity, Gemini and Google AI name your brand, its rivals and sources.

已验证Streamable HTTP可网页运行网页搜索与抓取数据与分析商业与电商

概览

AI 生成的概览

Firemark 是一个远程 MCP 服务器,让助手读取 AI 可见度追踪结果:ChatGPT、Perplexity、Gemini 和 Google AI 提及品牌、竞品和被引来源的频率。

可以做什么
  • 读取品牌结果:列出品牌、总体覆盖、各引擎计数、竞品排名与情感、被引来源、提示词与完整回答、运行记录和变化摘要。
  • 找出品牌失利之处:竞品被提及而品牌缺席的缺口、这些情况下被引用的网站、没有品牌页面出现的 ChatGPT 扇出查询,以及数字公关投放。
  • 设置追踪:进行五步访谈生成提示词草稿篮、保存草稿,并提出提示词或竞品变更建议。
  • 编辑品牌(仅在授予可选权限时):审阅草稿与提案,更新品牌信息、竞品、提示词、分组、网站类型和投放,并启动首次或重试运行。
  • 以 CSV 下载链接导出结果。
适用场景
当你追踪品牌在 AI 回答中的可见度,希望助手在对话中汇总覆盖情况、与竞品对比、发现内容缺口,或起草提示词与竞品变更时使用。
运行要求
需要一个至少含一个品牌的 Firemark 账户;MCP 服务器包含在所有套餐中,起价每月 29 美元。连接远程 Streamable HTTP 端点,并通过 OAuth 2.1 加 PKCE 登录;无需 API 密钥,本机不运行任何程序。
安装前请注意
登录会授予读取结果以及起草和提出变更的权限;批准和编辑品牌是单独权限,默认未勾选。编辑类工具会修改提示词、竞品、计划和投放,但草稿与提案需成员确认后才生效。部分随附技能可能提出用助手自带的网页工具读取你的页面、站点地图以及 AI 回答引用的页面(包括竞品的公开页面),并会先征求你的同意。只使用官方服务器地址。

安装

在 SourceWeft 中

  1. 打开 控制台中的 Firemark,将其添加到工作区。
  2. 连接 所需的 凭据。它们会被私密存储,绝不会在公开页面上展示。
  3. 为需要使用其工具的对话启用该服务。

Web executable,通过 Streamable HTTP。 远程服务在工作区中配置后即可从网页运行时运行。

交给智能体安装

把这段提示词粘贴到 Claude Code、Codex、Cursor 等智能体,或 SourceWeft 对话中。智能体会读取该服务器的安装指南,先向你展示发布者、端点和登录方式,经你同意后再把它添加到你的客户端。密钥不会写进任何命令。

阅读 https://sourceweft.com/mcp/com.tryfiremark%2Ffiremark/install.md,并按其中说明把这个 MCP 服务器接入我的客户端。修改任何配置之前,先告诉我它的发布者、端点和登录方式,等我确认。不要把 API 密钥写进命令或我会提交的文件;问我密钥应从哪里获取。

查看智能体所依据的安装指南

其他 MCP 客户端

把它添加到你客户端的 mcpServers 配置中。 客户端会在首次连接时完成 OAuth 登录。

{
  "mcpServers": {
    "firemark": {
      "type": "http",
      "url": "https://app.tryfiremark.com/mcp"
    }
  }
}

README

[Firemark]

Firemark MCP server

Firemark is the AI visibility tracker built to be run from your AI assistant. Its MCP server is on every plan from $29, so your assistant can read the brand's rank, competitors, cited sources and ChatGPT fan-out queries, set up the prompt set by interview, and, with your permission, approve and edit your brands.

Firemark asks a fixed set of buyer questions to ChatGPT, Perplexity, Gemini, Google AI Overviews and Google AI Mode on a schedule, and shows which brands each answer names and which sites it cites.

This repository connects Firemark to Cursor, Claude, Claude Code, ChatGPT, Gemini CLI, Codex, VS Code, Windsurf, Cline, Zed and other MCP clients. It is a Cursor plugin, a Claude Code plugin marketplace and a Gemini CLI extension in one. The plugin also carries ten skills: one on using Firemark's tools well, and nine workflows that turn the results into page changes (see Skills). Your assistant can then read your results, find where competitors are named and you are not, and draft prompt and competitor changes for you to confirm.

https://app.tryfiremark.com/mcp

That is the only official address. It is a remote MCP server (Streamable HTTP). You sign in with your Firemark account (OAuth 2.1 with PKCE). There is no API key, and nothing runs on your machine.

You need a Firemark account with at least one brand: tryfiremark.com. The MCP server is included on every plan, from $29 a month.

Connect

Cursor

Install the Firemark plugin from the Cursor Marketplace, or add the server yourself:

  1. Open Customize in Cursor and choose New MCP Server. Cursor opens mcp.json.
  2. Add Firemark and save:
    json
    {  "mcpServers": {    "firemark": {      "url": "https://app.tryfiremark.com/mcp"    }  }}
  3. Click Connect beside firemark, sign in to Firemark and choose Allow.

Cursor CLI: put the same entry in ~/.cursor/mcp.json, then run:

sh
agent mcp login firemark

The CLI keeps the sign-in per folder, so run agent from the folder you signed in from.

Claude (claude.ai and Claude Desktop)

  1. Open Customize, then Connectors.
  2. Click + Add, then Add custom connector. Name it Firemark and paste the server address.
  3. Click Connect, sign in to Firemark and choose Allow.

On Team and Enterprise plans, an owner adds the connector first under Organization settings.

Claude Code

Install the Firemark plugin from this repository:

/plugin install firemark --marketplace Firemark-ai/firemark-mcp

On Claude Code versions without that form, run /plugin marketplace add Firemark-ai/firemark-mcp, then /plugin install firemark@firemark. Or add the server alone:

sh
claude mcp add --transport http firemark https://app.tryfiremark.com/mcp

Then run /mcp in Claude Code, choose firemark and sign in.

ChatGPT

  1. On chatgpt.com, open Plugins, click +, then Add custom MCP server.
  2. Name it Firemark, paste the server address under Connection and choose OAuth.
  3. Create it as a plugin, then install it: sign in to Firemark and choose Allow.

ChatGPT adds custom MCP servers on the web only. On Business, Enterprise and Edu plans, a workspace admin has to allow plugins with MCP servers first.

VS Code

Add to .vscode/mcp.json, then start the server from the file and sign in:

json
{  "servers": {    "firemark": {      "type": "http",      "url": "https://app.tryfiremark.com/mcp"    }  }}

Gemini CLI

sh
gemini extensions install https://github.com/Firemark-ai/firemark-mcp

Then start Gemini CLI and run /mcp auth firemark to sign in.

OpenAI Codex CLI

sh
codex mcp add firemark --url https://app.tryfiremark.com/mcpcodex mcp login firemark

Or add it to ~/.codex/config.toml:

toml
[mcp_servers.firemark]url = "https://app.tryfiremark.com/mcp"

Windsurf

Add to Windsurf's mcp_config.json (Windsurf Settings, Cascade, MCP Servers, View raw config) and refresh. Windsurf opens the sign-in on first use.

json
{  "mcpServers": {    "firemark": {      "serverUrl": "https://app.tryfiremark.com/mcp"    }  }}

Cline

Add to cline_mcp_settings.json (MCP Servers, then Configure):

json
{  "mcpServers": {    "firemark": {      "type": "streamableHttp",      "url": "https://app.tryfiremark.com/mcp"    }  }}

Zed

Add to Zed's settings.json. Zed asks you to sign in the first time it connects.

json
{  "context_servers": {    "firemark": {      "url": "https://app.tryfiremark.com/mcp"    }  }}

Other MCP clients

Add https://app.tryfiremark.com/mcp as a remote (Streamable HTTP) MCP server with OAuth. The client opens Firemark's sign-in, then a consent page that asks what it may do.

Skills

The plugin carries ten skills. Claude Code and Cursor load them automatically, and each one turns on when your question matches it. In Claude Code you can also run them as slash commands, such as /firemark:firemark-weekly-check.

SkillWhat it does
firemarkWhich tool answers which question, and how to report Firemark's figures exactly
firemark-weekly-checkWhat changed since the last run and the one change to make this week; can run as a scheduled task
firemark-page-auditChecks one of your pages against the questions it should win, and lists what to add
firemark-write-the-fixDrafts the section a page is missing, in its own tone, with [CHECK] on any fact to confirm; also checks a draft before you publish
firemark-google-ai-citationsWhat Google AI Overviews, AI Mode and Gemini cite instead of you, and what those pages have
firemark-learn-from-winnersFor questions a competitor wins, the pages AI cites and how they differ from yours
firemark-fix-ai-factsWrong or vague statements about your brand in AI answers, and the sentence that would fix each
firemark-mentions-to-linksAnswers that name you without linking to you, and pages ChatGPT found but passed over
firemark-content-gapsChatGPT searches where none of your pages come back, grouped into page ideas, and what gets cited when you lose
firemark-comparison-pagesYou against one competitor, question by question, and an outline of the comparison page

What the skills do beyond calling Firemark: some offer to read web pages: your own pages, your sitemap, and the pages AI answers cite (including competitors' public pages). They ask you first, and read a page with the assistant's own web tool only if you agree; otherwise they ask you to paste it. firemark-write-the-fix and firemark-fix-ai-facts draft copy for you to review. No skill publishes anything, edits your site, or sends data anywhere except Firemark's server. Changes to a brand go through Firemark's own confirmation, as below.

Permissions

The consent page shows three permissions. You can revoke any connection at once on the Connect the MCP page in the app.

Permission (as the consent page names it)What it allowsOn the consent page
Read your resultsThe counts, verdicts and short snippets your screens show, the pages answers cite, and the full text of one answer at a time, on requestGranted when you choose Allow
Draft, propose and start a first runThe setup interview and a saved draft, proposals to add or remove prompts and competitors, a confirmed brand's first run, or one retry after a failed run. A draft or proposal takes effect only once a member confirms it.Granted when you choose Allow
Approve and edit your brandsApprove or discard drafts and proposals, and edit a brand's prompts, competitors, groups, schedule and emailsOnly if you tick its box (unticked to start)

No permission lets it start any other run, remove a brand, change a plan, see billing, or manage members or API keys.

Tools

Read your results

ToolWhat it returns
list_brandsThe brands you can read, across your accounts, with each brand's state
get_overviewThe headline of a brand's latest run, or a period's: answers naming the brand out of answers collected
get_enginesEach engine's answers naming the brand, out of its base
get_competitorsEach tracked competitor against the brand, by engine, with rank and sentiment, and how often each is mentioned first
get_sourcesThe sites and pages the answers cite, split by whether they name the brand or a competitor
list_promptsThe brand's prompts with one verdict per engine (named and linked, named, linked, absent), and the brand's place in each answer (order)
get_promptOne prompt in detail: each engine's verdict, position, sentiment, competitors and cited URLs
get_answerOne engine's answer to one prompt, word for word
get_runsThe run log: dates, status, coverage by engine and answers collected
get_gapsWhere the brand loses: answers that name a competitor and not the brand, ordered for action
get_source_gapsThe sites cited when a competitor is named and the brand is not
get_fanoutsChatGPT's fan-out queries, and whether the brand's own pages came back to them
get_changesWhat changed since a run, in one digest
get_whats_newProduct news: what changed in Firemark
get_export_linkA link to download a brand's CSV files, or a period's file
get_digital_prThe brand's Digital PR placements (articles on other sites that feature it) and how often AI answers cite them

Draft, propose and start a first run

ToolWhat it does
interview_startStarts or resumes a five-step interview that ends in a draft prompt basket
interview_submit_stepStores the answers to one step and returns the next
interview_get_basketReads the proposed prompts
interview_save_draftSaves the draft and returns the link where a member confirms it
propose_changesProposes prompts and competitors to add or remove on a running brand
start_runStarts a confirmed brand's first run, or one retry after a failed run

Approve and edit your brands (only if you tick it on the consent page)

ToolWhat it does
get_setupReads a brand's setup as it stands now
review_draftConfirms or discards a new brand's draft
review_proposalApproves or discards a brand's open proposal
update_brandEdits a brand's details, schedule or emails
update_competitorsAdds, removes or edits tracked competitors
update_promptsAdds, removes or edits prompts
manage_groupsCreates, renames or deletes prompt groups
set_site_typeSets the type of a cited site
update_placementsAdds or removes Digital PR placements, using only the article links you give

Example questions

  • Which brands do I track?
  • How often does ChatGPT name us, out of how many answers?
  • Which competitors are named where we are not?
  • Which sites do the answers cite when a competitor is named and we are not?
  • Which fan-out queries does ChatGPT run where none of our pages come back?
  • What changed since last week's run?
  • Help me set up tracking for a new brand.

Security

  • Sign-in is OAuth 2.1 with PKCE. Your assistant never sees your password.
  • Access tokens last one hour. Refresh tokens last 30 days and rotate.
  • Every call is checked against your account membership. A brand you are not a member of answers as if it does not exist.
  • Answer text and other quoted content come back marked as data, not instructions.
  • Use only https://app.tryfiremark.com/mcp. Firemark does not publish any other server address or package.

Report a security issue: see SECURITY.md.

Links

License

MIT

来源:README.md,提交 8aacdc0

工具

0
该服务器需要登录,因此 SourceWeft 无法在没有账号的情况下读取其工具,也未收录。已连接账号所能看到的内容可能因账号及其权限而异。

版本历史

3
  1. v1.6.0最新Nov 7, 2026
  2. v1.1.0Nov 4, 2026
  3. v1.0.0Oct 14, 2026

评分

已满足适用检查项的 85%

该评分衡量此收录条目的完整程度,依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价,也不表示服务器安全或当前可用:可连通性是带日期的检查结果,评分也不包含安装量或用户评价。

  • 近 30 天内可连通必需25 分
  • 已声明认证方式必需10 分
  • 已列出工具15 分

    请让服务器在无凭据时响应 tools/list,或发布可在沙箱中无凭据启动的软件包。

  • 描述必需15 分
  • 图标10 分
  • 文档15 分
  • 许可证10 分