Skip to main content
Glama
confluence-mcp-server
by qihaze123

confluence-mcp-server

一个面向 Confluence Cloud / Server / Data Center 的 MCP 服务,提供页面搜索、CQL 查询、读取、创建、更新、附件上传和用户身份查询能力。

功能

  • confluence_search_pages: 按关键词搜索页面

  • confluence_execute_cql_search: 执行原生 CQL 查询

  • confluence_get_page: 按页面 ID 获取内容(storage 格式)

  • confluence_get_page_outline: 获取页面标题目录,适合低 token 导航

  • confluence_get_page_section: 按标题读取单个 section,避免把整页正文返回给模型

  • confluence_get_page_anchor_block: 按 Anchor 起止边界读取块内容

  • confluence_create_page: 创建页面(支持可选 parentId)

  • confluence_update_page: 更新页面内容并自动递增版本

  • confluence_update_page_section: 按标题更新单个 section,由 MCP 服务端完成整页替换与提交

  • confluence_preview_page_section_update: 预览按标题更新的命中范围,返回旧内容和确认 hash

  • confluence_update_page_section_confirmed: 带 hash 确认提交按标题更新,避免 preview 后内容已变化

  • confluence_add_anchor_block_to_section: 给一个标题 section 自动加上起止 Anchor,便于后续稳定更新

  • confluence_preview_page_anchor_block_update: 预览按 Anchor 范围更新的命中块,返回旧内容和确认 hash

  • confluence_update_page_anchor_block_confirmed: 带 hash 确认提交 Anchor 范围更新

  • confluence_stage_page_update: 暂存整页改动到 MCP 服务端内存,不提交到 Confluence

  • confluence_stage_page_section_update: 暂存标题 section 改动到 MCP 服务端内存,不提交到 Confluence

  • confluence_stage_page_anchor_block_update: 暂存 Anchor 块改动到 MCP 服务端内存,不提交到 Confluence

  • confluence_list_pending_page_updates: 列出当前 MCP 进程里的未提交 wiki 改动

  • confluence_get_pending_page_update: 查看某个未提交改动的 old/new/diff 内容

  • confluence_commit_pending_page_update: 校验 hash 后把某个未提交改动真正提交到 Confluence

  • confluence_discard_pending_page_update: 丢弃某个未提交改动

  • confluence_upload_attachment: 上传页面附件(支持本地文件路径或 base64,同名附件默认更新为新版本)

  • confluence_get_current_user: 获取当前认证用户(whoami)

Related MCP server: confluence-mcp

环境要求

  • Node.js 18+

  • 可访问的 Confluence Cloud / Server / Data Center

快速开始

npm install
npm run build
npm start

通过 npx 使用(推荐)

发布到 npm 后,可在 MCP 客户端中直接使用 npx 启动,无需手动克隆仓库:

{
  "mcpServers": {
    "confluence": {
      "command": "npx",
      "args": ["-y", "confluence-mcp-server"],
      "env": {
        "CONF_MODE": "server",
        "CONF_BASE_URL": "https://confluence.example.com",
        "CONF_USERNAME": "your-username",
        "CONF_TOKEN": "your-token",
        "CONF_DEFAULT_SPACE": "DOC"
      }
    }
  }
}

也可以不在客户端配置里写连接信息,先只启动 MCP 服务:

{
  "mcpServers": {
    "confluence": {
      "command": "npx",
      "args": ["-y", "confluence-mcp-server"]
    }
  }
}

然后在对话里提供连接参数,让模型先调用 confluence_configure_connection:

{
  "baseUrl": "https://confluence.example.com",
  "mode": "server",
  "authMode": "auto",
  "username": "your-username",
  "token": "your-token",
  "defaultSpace": "DOC"
}

该配置只保存在当前 MCP 服务进程内存中,重启后会丢失。长期使用仍建议放在 MCP 客户端的环境变量配置里,避免在对话上下文中暴露密钥。

如果当前进程已经有连接(例如通过环境变量连到了旧地址),也可以只切换部分字段。比如只把地址从一台 Confluence 切到另一台、账号密码不变:

{
  "baseUrl": "http://192.168.40.82:8090"
}

未传的 mode、认证信息和 defaultSpace 会继承当前连接;如果显式传入 token / password / username / authMode,则会重新计算认证。

环境变量

  • CONF_BASE_URL: Confluence 基础地址,例如 https://confluence.example.com

  • CONF_MODE: 部署模式,cloud 或 server(默认 server)

  • CONF_AUTH_MODE: 认证模式,auto / basic / bearer(默认 auto)

  • CONF_USERNAME: 登录用户名(Cloud 必填;Server 在 basic 模式或使用密码时必填)

  • CONF_PASSWORD: 用户密码(与 CONF_TOKEN 组合按模式使用)

  • CONF_TOKEN: 访问令牌(Cloud 下作为 API Token;Server 下默认走 Bearer)

  • CONF_DEFAULT_SPACE: 默认空间 Key(可选)

如果未设置这些环境变量,服务会以“未配置连接”的状态启动;此时先调用 confluence_configure_connection 即可。可用 confluence_get_connection_status 查看当前连接状态,该工具不会返回 token/password。

MCP 客户端配置示例

{
  "mcpServers": {
    "confluence": {
      "command": "node",
      "args": ["dist/index.js"],
      "env": {
        "CONF_MODE": "server",
        "CONF_BASE_URL": "https://confluence.example.com",
        "CONF_USERNAME": "your-username",
        "CONF_TOKEN": "your-token",
        "CONF_DEFAULT_SPACE": "DOC"
      }
    }
  }
}

说明

  • 该项目当前聚焦 Confluence 能力,不包含 Jira 工具。

  • Cloud 模式固定使用 Basic(CONF_USERNAME + CONF_TOKEN/CONF_PASSWORD)。

  • Server 模式可用 Bearer 或 Basic(CONF_AUTH_MODE=auto 时优先 Bearer)。

  • Cloud 模式 API 基础路径为 /wiki/rest/api,Server 模式为 /rest/api。

  • 上传图片后,confluence_upload_attachment 会返回 storageImageMarkup,可把它拼到 confluence_update_page 的 bodyStorageValue 中展示图片。

  • Confluence 原生更新仍是整页版本化 PUT;confluence_update_page_section 只是把“整页读取 + 局部替换 + 整页提交”放在 MCP 服务端完成,从而显著减少模型侧 tokens。

  • 两阶段更新不会减少 Confluence 侧请求次数,但会显著降低误改风险:preview 返回旧内容和 expectedCurrentHash,confirmed update 会重新拉取页面并校验 hash 一致后才提交。

  • 暂存更新工具会把改动保存在当前 MCP 服务端进程内存中,不会调用 Confluence PUT;重启 MCP 服务后这些未提交草稿会丢失。

  • Anchor 模式使用 Confluence Anchor 宏作为隐形边界,适合长期自动维护固定区域;普通阅读模式下通常不会显示这些锚点。

低 Token 推荐流程

  1. 先用 confluence_search_pages 或 confluence_execute_cql_search 定位页面。

  2. 再用 confluence_get_page_outline 获取标题目录,不直接读整页正文。

  3. 用 confluence_get_page_section 按 heading 拉取需要处理的 section。

  4. 修改完成后,用 confluence_update_page_section 只提交该 section 的 storage 内容。

如果页面里存在重复标题,可以配合 occurrence 指定第几个同名标题;matchMode=contains 可用于模糊匹配标题。

更安全的两阶段更新

  1. 先调用 confluence_preview_page_section_update。

  2. 检查返回的 oldStorageValue 和 newStorageValue 是否符合预期。

  3. 把返回的 expectedCurrentHash 原样传给 confluence_update_page_section_confirmed。

  4. 服务端会重新拉取页面并校验 hash,一致才真正提交更新。

未提交改动流程

如果希望 AI 先改 wiki、你检查后再决定是否真正保存,可以使用暂存工具:

  1. 用 confluence_stage_page_section_update、confluence_stage_page_anchor_block_update 或 confluence_stage_page_update 创建未提交草稿。

  2. 检查返回的 oldStorageValue、newStorageValue、diffStorageValue 和 draftId。

  3. 需要稍后查看时,用 confluence_list_pending_page_updates 或 confluence_get_pending_page_update。

  4. 确认保存时,用 confluence_commit_pending_page_update 提交;服务端会重新拉取页面并校验 expectedCurrentHash,一致才写入 Confluence。

  5. 不想保存时,用 confluence_discard_pending_page_update 丢弃。

这些草稿只存在于当前 MCP 服务端内存,不会出现在 Confluence 页面历史里,也不会跨服务重启持久化。

更稳定的 Anchor 更新

  1. 先用 confluence_add_anchor_block_to_section 给目标 section 加上 startAnchor / endAnchor。

  2. 后续读取时用 confluence_get_page_anchor_block。

  3. 更新时先用 confluence_preview_page_anchor_block_update,确认后再调用 confluence_update_page_anchor_block_confirmed。

Anchor 名称建议使用业务语义化命名,例如 risk-summary-start、risk-summary-end,避免依赖标题文本本身。

Available Tools

23 tools
confluence_add_anchor_block_to_sectionA

Insert invisible start/end anchor macros around a heading-based section so later updates can target a stable block instead of relying on heading matching.

Inputs: pageId (required), heading (required), startAnchor (required), endAnchor (required), +5 more.

confluence_commit_pending_page_updateA

Commit one staged page update to Confluence by draftId after re-reading and hash-checking the current content.

Inputs: draftId (required), minorEdit, message.

confluence_create_pageC

Create a Confluence page. Supports optional parentId. Returns unified fields.

Inputs: title (required), bodyStorageValue (required), spaceKey, parentId.

confluence_discard_pending_page_updateA

Discard one staged page update by draftId without writing it to Confluence.

Inputs: draftId (required).

confluence_get_current_userA

Get current authenticated Confluence user (whoami).

Inputs: no named parameters.

confluence_get_pageB

Get a Confluence page by ID. Returns unified fields and bodyStorageValue.

Inputs: pageId (required), expand.

confluence_get_page_anchor_blockA

Get a page block between two invisible anchor macros. Returns only the matched block to save tokens.

Inputs: pageId (required), startAnchor (required), endAnchor (required).

confluence_get_page_outlineA

Get a page outline by parsing headings from body.storage. Returns unified fields plus headings[] for low-token navigation.

Inputs: pageId (required).

confluence_get_page_sectionA

Get a single page section by heading. The server fetches full storage internally but only returns the matched section to save tokens.

Inputs: pageId (required), heading (required), occurrence, includeHeading, +1 more.

confluence_get_pending_page_updateA

Get one staged page update by draftId. The draft has not been written to Confluence.

Inputs: draftId (required).

confluence_list_pending_page_updatesA

List staged page updates kept in MCP server memory. These drafts have not been written to Confluence.

Inputs: pageId.

confluence_preview_page_anchor_block_updateA

Preview an anchor-based block update. Returns the current block, the proposed replacement, and a hash that must be echoed to the confirmed update call.

Inputs: pageId (required), startAnchor (required), endAnchor (required), blockStorageValue (required).

confluence_preview_page_section_updateA

Preview a heading-based section update. Returns the currently matched section, the proposed replacement, and a hash that must be echoed to the confirmed update call.

Inputs: pageId (required), heading (required), sectionStorageValue (required), occurrence, +2 more.

confluence_search_pagesB

Search Confluence pages by keyword. Returns unified fields: id, type, title, spaceKey, url, version.

Inputs: query, spaceKey, limit.

confluence_stage_page_anchor_block_updateA

Stage an anchor-based block update in MCP server memory without writing to Confluence. Returns a draftId plus old/new/diff content for review.

Inputs: pageId (required), startAnchor (required), endAnchor (required), blockStorageValue (required), +1 more.

confluence_stage_page_section_updateA

Stage a heading-based section update in MCP server memory without writing to Confluence. Returns a draftId plus old/new/diff content for review.

Inputs: pageId (required), heading (required), sectionStorageValue (required), occurrence, +3 more.

confluence_stage_page_updateA

Stage a full-page update in MCP server memory without writing to Confluence. Use list/get to review, commit to save, or discard to drop it.

Inputs: pageId (required), bodyStorageValue (required), title.

confluence_update_pageA

Update a Confluence page content. Auto increments version. Returns unified fields.

Inputs: pageId (required), bodyStorageValue (required), title, minorEdit, +1 more.

confluence_update_page_anchor_block_confirmedA

Confirm an anchor-based block update after preview. The server re-reads the page and only updates when the preview hash still matches current content.

Inputs: pageId (required), startAnchor (required), endAnchor (required), blockStorageValue (required), +4 more.

confluence_update_page_sectionA

Update one page section by heading. The server fetches full storage internally, replaces the matched range, and writes back the whole page so the model only sends the section content.

Inputs: pageId (required), heading (required), sectionStorageValue (required), occurrence, +5 more.

confluence_update_page_section_confirmedA

Confirm a heading-based section update after preview. The server re-reads the page and only updates when the preview hash still matches current content.

Inputs: pageId (required), heading (required), sectionStorageValue (required), expectedCurrentHash (required), +6 more.

confluence_upload_attachmentA

Upload an attachment to a Confluence page. Supports local filePath or base64Data; same-name files are updated by default. Returns attachment fields and image storage markup when applicable.

Inputs: pageId (required), filePath, base64Data, fileName, +4 more.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.7/5.0

    Scored across 23 tools

    Disambiguation5/5

    Each tool has a unique purpose clearly indicated by its name and description. There is no overlap between tools like create_page, update_page, stage_page_update, etc., as they target different actions (create, read, update, stage, commit) and granularities (page, section, anchor block).

    Naming Consistency5/5

    All tools follow the consistent pattern 'confluence_verb_noun' (e.g., confluence_create_page, confluence_get_page_section). The verbs are specific and the nouns clearly indicate the target, forming a predictable and uniform naming scheme.

    Tool Count4/5

    With 23 tools, the server is on the higher side but justifiable for the domain. The staging/preview/commit workflow for sections and anchor blocks adds necessary tools for safe editing, but the count is slightly above the ideal range for a focused server.

    Completeness3/5

    The tool surface covers page creation, reading (with sections/blocks), updating, and attachment upload, plus search and user info. However, a delete page tool is missing, and updating page titles is not explicitly supported, leaving notable gaps in lifecycle coverage.

    Maintenance

    ActivityInactive
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers