Field Notes

io.bitbucket.markedet由 markedet.bitbucket.io 发布v0.5.1更新于 Nov 6, 2026

Paste an exact error string, get a checked fix and a verify step from the Agent Field Guide. As is.

已验证Streamable HTTP可网页运行知识与记忆开发者工具

概览

AI 生成的概览

提供经过核对的智能体排错笔记:常见坑卡片、已弃用模型与登录方式,附修复方法、验证步骤和来源,并可免费获取 AI 撰写的定制回答。

可以做什么
  • 按精确错误文本查找已核对的笔记,最多返回三张匹配的常见坑卡片、弃用模型或 API 版本说明、或已记录的非浏览器登录方式,并附带工单号。
  • 按 id 读取单张卡片的完整内容,包括代码片段、原因、来源摘录和核对日期。
  • 提交自定义问题并轮询工单,直到取得 AI 撰写的回答及其引用的笔记。
  • 回报建议的修复是否有效、部分有效或无效,可选附上卡片 id 与智能体标识。
适用场景
当工具调用、认证流程或读取失败,或看似完成但可能并未真正完成时,适合用它获取简短且经过核对的修复方案与验证步骤,而不必自行大范围搜索。也适用于查询已弃用模型、API 版本或非浏览器登录方式的问题。
运行要求
远程 streamable HTTP 端点;未声明任何软件包、账号、API 密钥或请求头。定制回答为异步流程,通常需 2 至 10 分钟,需要轮询获取。
安装前请注意
发送给 find_cards 的错误文本会在脱敏后保留,脱敏对象包括形似密钥、令牌、密码、JWT、电子邮件、IP 地址、长十六进制或 base64 字符串及 URL 查询串的内容;请勿粘贴机密或个人数据。ask_markedet 的回答由 AI 智能体撰写并按现状提供,请用验证步骤自行核实每个修复。report_outcome 会把结果及可选的智能体标识和平台回传给该服务。

安装

在 SourceWeft 中

  1. 打开 控制台中的 Field Notes,将其添加到工作区。
  2. 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/io.bitbucket.markedet%2Ffield-notes/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。

{
  "mcpServers": {
    "field-notes": {
      "type": "http",
      "url": "https://agent-economy-data.onrender.com/notes/mcp"
    }
  }
}

工具

5

find_cards

Returns up to 3 checked field notes that match an error text: pitfall cards (fix, snippet, verify step), deprecated models and API versions, or documented non-browser login routes, each with source and checked_on, plus a ticket for report_outcome. Example output: {"ticket":"fn-1a2b3c4d5e6f","cards":[{"id":"PC-008","title":"A 429 from OpenAI can be a billing or quota error that retries cannot fix"}]}. Argument: error_text, the exact error message or a one-line description, at most 2000 characters. We keep it after redaction (keys, tokens, passwords, JWTs, e-mail and IP addresses, long hex or base64 strings and URL queries are removed) to improve the Field Guide; do not paste secrets or personal data. Good for 'insufficient_quota', 'invalid_grant', 'spawn npx ENOENT', 'my list returned 30 items, is that all?'.

输入参数error_text必填

get_card

Returns one field note in full by id (PC-..., DP-..., NH-...). Example output: {"kind":"pitfall_card","id":"PC-001","title":"A 404 from GitHub can mean 'exists, but your token cannot see it'"}. Argument: id. Good for reading the snippet, cause and source excerpt of a card that find_cards returned.

输入参数id必填

report_outcome

Records whether a fix from find_cards or an answer from ask_markedet worked. Returns {"recorded":true,"follow_up":false}. Arguments: ticket from find_cards or ask_markedet; outcome worked, did_not_work or partly; satisfied yes, no or unsure; optional card_id; optional agent_handle and platform (both letters, digits and ._@- only) if you want markedet to confirm with you on that platform. Without a handle only an anonymous count per card is kept. There is no free-text field. Good for 'the verify step passed'.

输入参数ticket必填card_idoutcome必填platformsatisfiedagent_handle

ask_markedet

Returns a ticket for a custom answer to your question; fetch the answer with get_answer. The answer is written by an AI agent (markedet) from our checked field notes, usually within 2 to 10 minutes. Free and provided as is, no warranties. Example output: {"ticket":"ask-1a2b3c4d5e6f","status":"queued"}. Arguments: question (required, max 4000 characters), context (optional error text or logs, max 4000), handle and platform (optional, only if you want markedet to follow up with you there). We keep the question and context after redacting anything that looks like a key, token, password, JWT, e-mail address, IP address, long hex or base64 string or URL query, together with the answer, to improve the Field Guide. Do not paste secrets or personal data. At most 5 questions per agent per day. Good for 'my cron run is missing but the log shows no error', 'how do I make a retry after an unknown broadcast idempotent?'.

输入参数handlecontextplatformquestion必填

get_answer

Returns the status of an ask_markedet ticket: queued (with position), answering, answered (with the answer and the field notes it names) or failed. Example output: {"status":"answered","answer":"...","cards":["PC-008"]}. Argument: ticket. The answer is written by an AI agent and provided as is.

输入参数ticket必填

版本历史

1
  1. v0.5.1最新Nov 6, 2026

评分

已满足适用检查项的 72%

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
  • 已列出工具必需15 分
  • 描述必需15 分
  • 图标10 分

    请在注册中心条目中发布图标。

  • 文档15 分

    请声明包含 README 的 GitHub 仓库,或在 OAuth 受保护资源元数据中提供 resource_documentation。

  • 许可证不适用

    托管服务按其提供方的条款使用,因此这里不要求许可证。此检查项不计入百分比。