Backlog MCP Server 1f04ad2c

MCP开发工具

提供方 nulab

开发工具 stdio · 本地进程 会话登录 社区 / 第三方 Claude DesktopClaudeCursorCline
查看接入方式

一个用于与Backlog API交互的模型上下文协议(MCP)服务器,提供通过AI代理管理Backlog项目、问题、Wiki页面等功能。

一句话选型结论

按资料字段整理

Backlog MCP Server 1f04ad2c 属于开发工具类 MCP 服务器,stdio · 本地进程,会话登录,可直接接入 Claude Desktop、Claude、Cursor 等,有可直接执行的安装命令。接入前先确认权限范围;涉及写入、删除或付款的操作,建议先在测试环境验证。

分类开发工具 传输方式stdio · 本地进程 鉴权会话登录 部署方式本地部署
GitHub 星标 185 仓库热度,无公开仓库时显示未标注
工具数 52 暴露给模型的工具数量
资源 / 提示词 0 / 0 资源与提示词模板数量
维护方 社区 / 第三方 核验等级 仅资料核对

工具与能力

开发工具
数量来自来源目录的标注,具体工具清单以官方文档为准
工具数量52暴露给模型可调用的工具个数
资源数量0可读取的资源条目
提示词数量0内置提示词模板数量

能力标签

支持客户端

适用场景

项目管理TypeScript混合部署问题跟踪Wiki管理Git集成API代理

安装与配置

stdio · 本地进程 会话登录
客户端在本机拉起子进程,数据不出本机
docker run -i --rm ghcr.io/nulab/backlog-mcp-server node build/index.js --export-translations

这条命令只是入口,具体启动参数请按官方文档补齐,不要直接照搬到别的服务器。

  1. 1
    准备运行环境

    stdio 服务在本地起进程,需要先装好 Docker,再按安装命令拉取依赖。

  2. 2
    安装服务

    复制上面的安装命令执行;命令里带占位符的位置,换成自己的密钥或路径。

  3. 3
    写入客户端配置

    在客户端的 MCP 配置里填好启动命令和参数,保存后完全退出客户端再打开,让新配置生效。

  4. 4
    验证是否可用

    连接成功后客户端会列出约 52 个工具,随便调一个只读工具确认能返回结果。

典型使用场景

3 类
按分类与标签整理的常见用法,具体以官方文档为准
代码与仓库操作

在对话里直接查代码、提 issue、走 PR,省掉来回切工具。

本地命令与脚本

把构建、测试、格式化这类重复命令交给模型执行。

接口联调与测试

让模型调你的接口、跑一遍用例,快速定位问题。

选择建议

选之前先看这几条
按条目字段给出的判断,不替你决定用不用

值得选的地方

  • 资料标注 52 个工具,覆盖面较广,能满足多种调用。
  • 有现成的安装命令,不用自己找入口,照着跑就能起服务。
  • 已适配 Claude Desktop、Claude、Cursor,换客户端不用重新研究配置。

先想清楚的地方

  • stdio 方式要求本地装好运行时(如 Node.js),换机器要重新装一遍。
  • 使用前要准备好会话登录对应的凭据,并按最小权限配置。
  • 版本更新与行为细节以官方仓库为准。

更适合

  • 需要在开发工具场景里让模型直接调用外部能力的人。
  • 已经在用 Claude Desktop、Claude 的用户,接入路径最短。
  • 想先快速试一下、能接受在本地或自有环境跑服务的人。

不太适合

  • 完全不想在本地装环境的人。
  • 不方便管理密钥或第三方授权的人。
  • 需要厂商级 SLA 和技术支持的生产环境。

同类对比

6 个同类
同分类里挑最接近的几条,字段以各自详情页为准
MCP 服务器工具数传输方式部署方式安装命令最后核验
Backlog MCP Server 1f04ad2c当前 52 stdio · 本地进程 本地部署 有 2026-05-17
Git 838e2ec1 12 stdio · 本地进程 本地部署 有 2026-05-17
Markitdown MCP 2c653a5f 1 stdio · 本地进程 本地部署 有 2026-05-17
JavaGuide 未标注 stdio · 本地进程 本地部署 有 2026-04-11
awesome-claude-skills 未标注 stdio · 本地进程 本地部署 有 2026-04-26
Context7 831acf40 未标注 stdio · 本地进程 本地部署 有 2026-05-17

服务器档案

开发工具
字段来自项目仓库与公开索引,中文说明由本站标注
提供方nulab上游标注的维护方 nulab
分类开发工具本站按上游分类展示
连接方式stdio · 本地进程客户端在本机拉起子进程,数据不出本机
鉴权方式会话登录远程服务建议用最小权限凭据
维护方社区 / 第三方核验等级 仅资料核对
包名 / 运行时未标注 · Docker本地安装时用到的包标识与运行时
部署类型本地部署上游给出的部署形态
最近核验2026-05-17上游最后核验时间

详细介绍

30.8 KB
来自项目仓库与上游目录的原文,未做删改

Backlog MCP Server

📘 日本語でのご利用ガイド

A Model Context Protocol (MCP) server for interacting with the Backlog API. This server provides tools for managing projects, issues, wiki pages, and more in Backlog through AI agents like Claude Desktop / Cline / Cursor etc.

Features

  • Project tools (create, read, update, delete)
  • Issue tracking and comments (create, update, delete, list)
  • Version/Milestone management (create, read, update, delete)
  • Wiki page support
  • Git repository and pull request tools
  • Notification tools
  • Field selection for optimized responses
  • Token limiting for large responses

Getting Started

Requirements

  • Docker
  • A Backlog account with API access
  • API key from your Backlog account

Option 1: Install via Docker

The easiest way to use this MCP server is through MCP configurations:

  1. Open MCP settings
  2. Navigate to the MCP configuration section
  3. Add the following configuration:```json
    {
    "mcpServers": {
    "backlog": {
    "command": "docker",
    "args": [
    "run",
    "--pull",
    "always",
    "-i",
    "--rm",
    "-e",
    "BACKLOG_DOMAIN",
    "-e",
    "BACKLOG_API_KEY",
    "ghcr.io/nulab/backlog-mcp-server"
    ],
    "env": {
    "BACKLOG_DOMAIN": "your-domain.backlog.com",
    "BACKLOG_API_KEY": "your-api-key"
    }
    }
    }
    }

Replace `your-domain.backlog.com` with your Backlog domain and `your-api-key` with your Backlog API key.

✅ If you cannot use --pull always, you can manually update the image using:```
docker pull ghcr.io/nulab/backlog-mcp-server:latest

Option 2: Install via npx

You can also run the server directly using npx without cloning the repository. This is a convenient way to run the server without a full installation.

  1. Open MCP settings
  2. Navigate to the MCP configuration section
  3. Add the following configuration:```json
    {
    "mcpServers": {
    "backlog": {
    "command": "npx",
    "args": ["backlog-mcp-server"],
    "env": {
    "BACKLOG_DOMAIN": "your-domain.backlog.com",
    "BACKLOG_API_KEY": "your-api-key"
    }
    }
    }
    }

Replace `your-domain.backlog.com` with your Backlog domain and `your-api-key` with your Backlog API key.

### Option 3: Manual Setup (Node.js)

1. Clone and install:```bash
   git clone https://github.com/nulab/backlog-mcp-server.git
   cd backlog-mcp-server
   pnpm install
   pnpm run build
   ```

2. Create `.env` from template and set required variables:```bash
cp .env.example .env

Set the following values in .env:

  • BACKLOG_DOMAIN=your-domain.backlog.com
  • BACKLOG_API_KEY=your-api-key
  1. Run locally:```bash
    pnpm run dev

4. Set your json to use as MCP```json
{
  "mcpServers": {
    "backlog": {
      "command": "node",
      "args": ["your-repository-location/build/index.js"],
      "env": {
        "BACKLOG_DOMAIN": "your-domain.backlog.com",
        "BACKLOG_API_KEY": "your-api-key"
      }
    }
  }
}

HTTP transport (Streamable HTTP)

By default the server uses stdio. To run the MCP Streamable HTTP transport instead (JSON-RPC over HTTP, same tools as stdio), start with --transport http or set MCP_TRANSPORT=http.```bash
pnpm run build
MCP_TRANSPORT=http MCP_HTTP_PORT=3333 node build/index.js


- **Endpoint:** `POST` (and `GET` for server-initiated streams) on `http://:` (default path `/mcp`).
- **Protocol:** MCP `2026-07-28`. The protocol is stateless: there is no `initialize` handshake and no `mcp-session-id` header. Clients send their metadata in `_meta` on every request and discover capabilities via `server/discover`. Streamable HTTP also requires the `Mcp-Method` header (and `Mcp-Name` on `tools/call`).
- **Backward compatibility:** Clients on `2025-11-25` and earlier are still served over the same endpoint, statelessly. Because no session is kept, the 2025 session operations (`GET` / `DELETE` with an `mcp-session-id`) answer `405`.
- **Security:** Default bind is `127.0.0.1`. On a bare loopback bind, `Host` and `Origin` are both validated against the localhost set (DNS rebinding protection). Behind a reverse proxy, set `--http-allowed-hosts` to the public hostname; that turns off the localhost `Origin` default, since a browser client's `Origin` is its own site and never this server's hostname. Add `--http-allowed-origins` to restrict which client origins may reach the server. Do not expose the HTTP port to untrusted networks without authentication and TLS; it allows full use of your Backlog API key via MCP tools.

Environment variables (CLI flags override when both are set):

| Variable                   | Description                                                                                                                                                                       |
| -------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `MCP_TRANSPORT`            | `stdio` (default) or `http`                                                                                                                                                       |
| `MCP_HTTP_HOST`            | Bind address (default `127.0.0.1`)                                                                                                                                                |
| `MCP_HTTP_PORT`            | Port (default `3333`)                                                                                                                                                             |
| `MCP_HTTP_PATH`            | URL path (default `/mcp`)                                                                                                                                                         |
| `MCP_HTTP_JSON_RESPONSE`   | `true` to prefer JSON responses over SSE (applies to `2026-07-28` clients only)                                                                                                   |
| `MCP_HTTP_ALLOWED_HOSTS`   | Comma-separated allowed `Host` hostnames (port-agnostic). Required when binding to `0.0.0.0`; also the escape hatch for a loopback bind behind a proxy (DNS rebinding protection) |
| `MCP_HTTP_ALLOWED_ORIGINS` | Comma-separated allowed `Origin` hostnames for browser-based clients. Defaults to the localhost set on a bare loopback bind, and to no `Origin` check otherwise                   |

### OAuth 2.0 Authentication (Remote MCP)

When exposing the MCP server over a network, you can enable OAuth 2.0 authentication so that each user authenticates with their own Backlog account instead of sharing a single API key.

The server implements the MCP Third-Party Authorization Flow by acting as both an OAuth authorization server (for MCP clients) and an OAuth client (for Backlog).

#### Prerequisites

1. Register an OAuth application in your Backlog space:
   - Go to your Backlog space → Personal Settings → Register Application
   - Set the **Redirect URI** to `/callback` (e.g., `https://mcp.example.com/callback`)
   - Note the **Client ID** and **Client Secret**

2. Set the following environment variables (in addition to `BACKLOG_DOMAIN`):

| Variable                      | Description                                                     |
| ----------------------------- | --------------------------------------------------------------- |
| `BACKLOG_OAUTH_CLIENT_ID`     | OAuth Client ID from your Backlog application                   |
| `BACKLOG_OAUTH_CLIENT_SECRET` | OAuth Client Secret from your Backlog application               |
| `MCP_SERVER_BASE_URL`         | Public URL of your MCP server (e.g., `https://mcp.example.com`) |

> **Note:** `BACKLOG_API_KEY` is **not required** when OAuth is enabled — each user authenticates with their own Backlog account.

#### Example```bash
BACKLOG_DOMAIN=your-space.backlog.com \
BACKLOG_OAUTH_CLIENT_ID=your-client-id \
BACKLOG_OAUTH_CLIENT_SECRET=your-client-secret \
MCP_SERVER_BASE_URL=https://mcp.example.com \
node build/index.js --transport http --http-host 0.0.0.0 --http-port 3333 \
  --http-allowed-hosts mcp.example.com

--http-allowed-hosts is required in practice when binding to 0.0.0.0: without it there is no DNS rebinding protection, and the server logs a warning at startup.

The server automatically exposes the following OAuth endpoints when OAuth is enabled:

Endpoint Description
GET /.well-known/oauth-authorization-server OAuth Authorization Server Metadata (RFC 8414)
GET /.well-known/oauth-protected-resource/mcp OAuth Protected Resource Metadata (RFC 9728)
POST /register Dynamic Client Registration (RFC 7591)
GET /authorize Authorization endpoint (redirects to Backlog OAuth)
GET /callback Backlog OAuth callback
POST /token Token endpoint (authorization code & refresh token)

MCP clients that support the MCP authorization specification will use these endpoints automatically.

POST /register restricts which redirect URIs a client may register. A loopback
URI (http://localhost, http://127.0.0.1, http://[::1]) is how an app running
on the user's machine receives the authorization code, and is accepted from a
client that declares "application_type": "native" — or, when the field is
absent, from one whose redirect URIs are all loopback. A client declaring
"application_type": "web", or mixing a remote https: URI with a loopback one
without declaring itself, is rejected with invalid_client_metadata.

> Limitations:
>
> - OAuth mode currently supports a single Backlog organization. It is not compatible with the multi-organization configuration.
> - Client registrations and tokens are stored in memory and will be lost on server restart.

Tool Configuration

You can selectively enable or disable specific toolsets using the --enable-toolsets command-line flag or the ENABLE_TOOLSETS environment variable. This allows better control over which tools are available to the AI agent and helps reduce context size.

Available Toolsets

The following toolsets are available (enabled by default when "all" is used):

Toolset Description
space Tools for managing Backlog space settings and general information
project Tools for managing projects, categories, custom fields, and issue types
issue Tools for managing issues and their comments, version milestones
wiki Tools for managing wiki pages
git Tools for managing Git repositories and pull requests
notifications Tools for managing user notifications
document Tools for viewing documents and document trees

Specifying Toolsets

You can control toolset activation in the following ways:

Using via CLI:```bash
--enable-toolsets space,project,issue


Or via environment variable:```
ENABLE_TOOLSETS="space,project,issue"

If all is specified, all available toolsets will be enabled. This is also the default behavior.

Using selective toolsets can be helpful if the toolset list is too large for your AI agent or if certain tools are causing performance issues. In such cases, disabling unused toolsets may improve stability.

> 🧩 Tip: project toolset is highly recommended, as many other tools rely on project data as an entry point.

Available Tools

Toolset: space

Tools for managing Backlog space settings and general information.

  • get_space: Returns information about the Backlog space.
  • get_users: Returns list of users in the Backlog space.
  • get_myself: Returns information about the authenticated user.

Toolset: project

Tools for managing projects, categories, custom fields, and issue types.

  • get_project_list: Returns list of projects.
  • add_project: Creates a new project.
  • get_project: Returns information about a specific project.
  • get_project_users: Returns list of users in a specific project.
  • update_project: Updates an existing project.

Toolset: issue

Tools for managing issues, their comments, and related items like priorities, categories, custom fields, issue types, resolutions, and watching lists.

  • get_issue: Returns information about a specific issue.
  • get_issue_attachment: Downloads one attachment of an issue. Returns it as image or embedded resource content, or as base64 with format: "base64".
  • get_issues: Returns list of issues.
  • count_issues: Returns count of issues.
  • add_issue: Creates a new issue in the specified project.
  • update_issue: Updates an existing issue.
  • delete_issue: Deletes an issue.
  • get_issue_comments: Returns list of comments for an issue.
  • add_issue_comment: Adds a comment to an issue.
  • update_issue_comment: Updates a comment on an issue.
  • get_related_issues: Returns list of issues related to a specific issue.
  • add_related_issue: Relates an issue to another issue.
  • remove_related_issue: Removes the relation between an issue and a related issue.
  • get_priorities: Returns list of priorities.
  • get_categories: Returns list of categories for a project.
  • add_category: Creates a new category for a project.
  • get_custom_fields: Returns list of custom fields for a project.
  • get_issue_types: Returns list of issue types for a project.
  • get_resolutions: Returns list of issue resolutions.
  • get_watching_list_items: Returns list of watching items for a user.
  • get_watching_list_count: Returns count of watching items for a user.
  • add_watching: Adds a new watch to an issue.
  • update_watching: Updates an existing watch note.
  • delete_watching: Deletes a watch from an issue.
  • mark_watching_as_read: Marks a watch as read.
  • get_version_milestone_list: Returns list of version milestones for a project.
  • add_version_milestone: Creates a new version milestone for a project.
  • update_version_milestone: Updates an existing version milestone.
  • delete_version_milestone: Deletes a version milestone.

Toolset: wiki

Tools for managing wiki pages.

  • get_wiki_pages: Returns list of Wiki pages.
  • get_wikis_count: Returns count of wiki pages in a project.
  • get_wiki: Returns information about a specific wiki page.
  • add_wiki: Creates a new wiki page.

Toolset: git

Tools for managing Git repositories and pull requests.

  • get_git_repositories: Returns list of Git repositories for a project.
  • get_git_repository: Returns information about a specific Git repository.
  • get_pull_requests: Returns list of pull requests for a repository.
  • get_pull_requests_count: Returns count of pull requests for a repository.
  • get_pull_request: Returns information about a specific pull request.
  • add_pull_request: Creates a new pull request.
  • update_pull_request: Updates an existing pull request.
  • get_pull_request_comments: Returns list of comments for a pull request.
  • add_pull_request_comment: Adds a comment to a pull request.
  • update_pull_request_comment: Updates a comment on a pull request.

Toolset: notifications

Tools for managing user notifications.

  • get_notifications: Returns list of notifications.
  • get_notifications_count: Returns count of notifications.
  • reset_unread_notification_count: Resets unread notification count.
  • mark_notification_as_read: Marks a notification as read.

Toolset: document

Tools for managing documents and document trees in Backlog projects.

  • get_document_tree: Returns the hierarchical tree of documents for a project, including folders and ne
  • get_documents: Returns a flat list of documents in a project or folder.
  • get_document_comments: Returns the comments on a document, including their replies.
  • get_document: Returns detailed information about a specific document, including metadata, content, an

Usage Examples

Once the MCP server is configured in AI agents, you can use the tools directly in your conversations. Here are some examples:

  • Listing Projects```
    Could you list all my Backlog projects?

- Creating a New Issue```
Create a new bug issue in the PROJECT-KEY project with high priority titled "Fix login page error"
  • Getting Project Details```
    Show me the details of the PROJECT-KEY project

- Working with Git Repositories```
List all Git repositories in the PROJECT-KEY project
  • Managing Pull Requests```
    Show me all open pull requests in the repository "repo-name" of PROJECT-KEY project
    ``````
    Create a new pull request from branch "feature/new-feature" to "main" in the repository "repo-name" of PROJECT-KEY project

- Watching Items```
Show me all items I'm watching

Overriding Tool Descriptions

You can override the descriptions of tools by creating a .backlog-mcp-serverrc.json file in your home directory.

Almost all of these strings are the tool and parameter descriptions the model reads when it decides which tool to call and how to fill in its arguments, so overriding them is a way to steer tool selection — for example to disambiguate two similar tools, or to add a rule your team follows — rather than a way to change the language of the answers you get. The model answers in whatever language you ask in, regardless of the language these descriptions are written in.

A small number of keys are validation error messages instead (for example PROJECT_ID_OR_KEY_REQUIRED). Those are returned in the tool result when a call is rejected, so they can reach you by way of the model's reply.

The file should contain a JSON object with the tool names as keys and the new descriptions as values.
For example:```json
{
"TOOL_ADD_ISSUE_COMMENT_DESCRIPTION": "An alternative description",
"TOOL_CREATE_PROJECT_DESCRIPTION": "Create a new project in Backlog"
}


When the server starts, it determines the final description for each tool based on the following priority:

1. Environment variables (e.g., `BACKLOG_MCP_TOOL_ADD_ISSUE_COMMENT_DESCRIPTION`)
2. Entries in `.backlog-mcp-serverrc.json` - Supported configuration file formats: .json, .yaml, .yml
3. Built-in defaults

Empty or non-string values are ignored at every level, and the built-in default is used instead.

Sample config:```json
{
  "mcpServers": {
    "backlog": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "-e",
        "BACKLOG_DOMAIN",
        "-e",
        "BACKLOG_API_KEY",
        "-v",
        "/yourcurrentdir/.backlog-mcp-serverrc.json:/root/.backlog-mcp-serverrc.json:ro",
        "ghcr.io/nulab/backlog-mcp-server"
      ],
      "env": {
        "BACKLOG_DOMAIN": "your-domain.backlog.com",
        "BACKLOG_API_KEY": "your-api-key"
      }
    }
  }
}

Exporting Current Descriptions

This prints every key that is resolved while the tool list is built, with its current value, including any customizations you have made. That covers all tool and parameter descriptions, and it is the practical way to discover key names.

It does not cover the validation error messages, because those keys are only resolved when a call is actually rejected. They are still overridable by the same rules; you just have to read them out of the source.

Example:```bash
docker run -i --rm ghcr.io/nulab/backlog-mcp-server node build/index.js --export-descriptions


or```bash
npx github:nulab/backlog-mcp-server --export-descriptions

Using Environment Variables

Alternatively, you can override tool descriptions via environment variables.

The environment variable names are based on the tool keys, prefixed with BACKLOGMCP and written in uppercase.

Example:
To override the TOOL_ADD_ISSUE_COMMENT_DESCRIPTION:```json
{
"mcpServers": {
"backlog": {
"command": "docker",
"args": [
"run",
"-i",
"--rm",
"-e", "BACKLOG_DOMAIN",
"-e", "BACKLOG_API_KEY",
"-e", "BACKLOG_MCP_TOOL_ADD_ISSUE_COMMENT_DESCRIPTION"
"ghcr.io/nulab/backlog-mcp-server"
],
"env": {
"BACKLOG_DOMAIN": "your-domain.backlog.com",
"BACKLOG_API_KEY": "your-api-key",
"BACKLOG_MCP_TOOL_ADD_ISSUE_COMMENT_DESCRIPTION": "An alternative description"
}
}
}
}


The server loads the config file synchronously at startup.

Environment variables always take precedence over the config file.

## Advanced Features

### Tool Name Prefixing

Add prefix to tool names with:```
--prefix backlog_

or via environment variable:```
PREFIX="backlog_"


This is especially useful if you're using multiple MCP servers or tools in the same environment and want to avoid name collisions. For example, get_project can become backlog_get_project to distinguish it from similarly named tools provided by other services.

### Response Optimization & Token Limits

#### Response Compaction

Tool results are compacted before they are returned. This is always on, and applies to every tool except `get_issue_attachment`, which returns the file itself:

- A user nested in a result (`createdUser`, `assignee`, `stars[].presenter`, a notification's `sender`, …) is reduced to `id`, `userId` and `name`. A user that *is* the result — `get_myself`, the items of `get_users` or `get_project_users` — is returned whole.
- A nested project (each activity's and notification's `project`) is reduced to `id`, `projectKey` and `name`, and a nested issue (a pull request's or notification's `issue`) to `id`, `issueKey`, `summary` and `status`. A project or issue that is the result — `get_project`, `get_issue`, the items of `get_project_list` or `get_issues` — is returned whole, and so is the issue of each item of `get_watching_list_items`.
- Nested entities (statuses, issue types, categories, milestones, custom field items, …) omit `color` and `displayOrder`, and nested milestones also omit `description`. Tools that list these themselves (`get_issue_types`, `get_categories`, `get_version_milestone_list`, …) return them whole.
- Fields that are `null`, an empty string or an empty array are omitted. `0`, `false` and `{}` are kept. This also applies to a field named in `fields`: an empty one is left out rather than returned as empty.
- JSON is returned without indentation.

#### Field Selection```
--optimize-response

Or environment variable:```
OPTIMIZE_RESPONSE=1


Tools that return a **list** then take an optional `fields` parameter: a list of
top-level field names from that tool's own result, published as an enum so a name
the tool does not have is rejected rather than ignored. Tools that return a single
record do not get it — the parameter costs schema on every session, and one record
has almost nothing to trim.```
get_project(projectIdOrKey: "PROJECT-KEY", fields: ["name", "key", "description"])

Omitting fields returns the whole result. Selection is one level deep: naming an
object or array field returns it whole.

Benefits:

  • Reduce response size by requesting only needed fields
  • Focus on specific data points
  • Improve performance for large responses

Token Limiting

Large responses are automatically limited to prevent exceeding token limits:

  • Default limit: 50,000 tokens
  • Configurable via MAX_TOKENS environment variable
  • Responses exceeding the limit are truncated with a message

You can change this using:```
MAX_TOKENS=10000


If a response exceeds the limit, it will be truncated with a warning.

> Note: This is a best-effort mitigation, not a guaranteed enforcement.

### Logging

The server logs to **stderr** (stdout carries the JSON-RPC stream on the stdio transport).

| Variable    | Description                                                                                                                                                                                             |
| ----------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `LOG_LEVEL` | `fatal`, `error`, `warn`, `info`, `debug`, `trace` or `silent`. Defaults to `error` when `NODE_ENV` is `production` — which is also the default when `NODE_ENV` is unset — and to `debug` otherwise. An unrecognised value is reported and the default is used. |

`NODE_ENV` still selects the output *format*: any value other than `production` switches to human-readable `pino-pretty` output when that package is available. Use `LOG_LEVEL`, not `NODE_ENV`, to change how much is logged, so that a deployment keeps structured JSON:

`pino-pretty` is a development dependency, so neither the published npm package nor the container image carries a copy. In those, logs are structured JSON whatever `NODE_ENV` says, and `LOG_LEVEL` is the only setting that changes the output.```
LOG_LEVEL=info node build/index.js --transport http

Full Custom Configuration Example

This section demonstrates advanced configuration using multiple environment variables. These are experimental features and may not be supported across all MCP clients. This is not part of the MCP standard specification and should be used with caution.```json
{
"mcpServers": {
"backlog": {
"command": "docker",
"args": [
"run",
"-i",
"--rm",
"-e",
"BACKLOG_DOMAIN",
"-e",
"BACKLOG_API_KEY",
"-e",
"MAX_TOKENS",
"-e",
"OPTIMIZE_RESPONSE",
"-e",
"PREFIX",
"-e",
"ENABLE_TOOLSETS",
"ghcr.io/nulab/backlog-mcp-server"
],
"env": {
"BACKLOG_DOMAIN": "your-domain.backlog.com",
"BACKLOG_API_KEY": "your-api-key",
"MAX_TOKENS": "10000",
"OPTIMIZE_RESPONSE": "1",
"PREFIX": "backlog_",
"ENABLE_TOOLSETS": "space,project,issue"
}
}
}
}


## Development

### Running Tests```bash
pnpm test

Adding New Tools

  1. Create a new file in src/tools/ following the pattern of existing tools
  2. Create a corresponding test file
  3. Add the new tool to src/tools/tools.ts
  4. Build and test your changes

Command Line Options

The server supports several command line options:

  • --transport stdio|http: MCP transport (default: stdio). Use http for Streamable HTTP.
  • --http-host, --http-port, --http-path: HTTP bind address, port, and path (defaults: 127.0.0.1, 3333, /mcp).
  • --http-json-response: Prefer JSON responses over SSE. Applies to 2026-07-28 clients only; the backward-compatible 2025-11-25 path is served with the SDK's default response shaping.
  • --http-allowed-hosts: Comma-separated allowed Host hostnames (port-agnostic). Needed when binding to all interfaces, or on a loopback bind behind a reverse proxy.
  • --http-allowed-origins: Comma-separated allowed Origin hostnames for browser-based clients. Defaults to the localhost set on a bare loopback bind, and to no Origin check otherwise.
  • --optimize-response: Add a fields parameter to each tool for selecting which result fields to return
  • --max-tokens=NUMBER: Set maximum token limit for responses
  • --prefix=STRING: Optional string prefix to prepend to all tool names (default: "")
  • --enable-toolsets: Specify which toolsets to enable (comma-separated or multiple arguments). Defaults to "all".
    Example: --enable-toolsets space,project or --enable-toolsets issue --enable-toolsets git
    Available toolsets: space, project, issue, wiki, git, notifications.

Example:```bash
node build/index.js --optimize-response --max-tokens=100000 --prefix="backlog_" --enable-toolsets space,issue


HTTP example:```bash
node build/index.js --transport http --http-port 3333 --http-path /mcp

Multi-Organization Support

This server can be configured to access multiple Backlog organizations from a single MCP server instance.

Configuration

Configure one env pair per organization and set a default organization:```bash
BACKLOG_DEFAULT_ORG=COMPANY_A
BACKLOG_ORG_COMPANY_A_DOMAIN=company-a.backlog.com
BACKLOG_ORG_COMPANY_A_API_KEY=your-company-a-api-key
BACKLOG_ORG_COMPANY_B_DOMAIN=company-b.backlog.com
BACKLOG_ORG_COMPANY_B_API_KEY=your-company-b-api-key


This works whether the variables come from a local `.env`, your shell environment, or an MCP client config `env` block.

Example MCP config:```json
{
  "env": {
    "BACKLOG_DEFAULT_ORG": "COMPANY_A",
    "BACKLOG_ORG_COMPANY_A_DOMAIN": "company-a.backlog.com",
    "BACKLOG_ORG_COMPANY_A_API_KEY": "your-company-a-api-key",
    "BACKLOG_ORG_COMPANY_B_DOMAIN": "company-b.backlog.com",
    "BACKLOG_ORG_COMPANY_B_API_KEY": "your-company-b-api-key"
  }
}

If no multi-organization env vars are set, the server falls back to the existing single-organization configuration:bash BACKLOG_DOMAIN=your-domain.backlog.com BACKLOG_API_KEY=your-api-key

Tool Usage

When multi-organization env vars are configured, all normal tools accept an optional organization input field. When provided, the tool call is routed to that Backlog organization.

…(正文过长,页面只展示前面部分,完整内容见仓库)

常见问题

8 问
答案由本站依据条目字段整理,不引用上游原文
Backlog MCP Server 1f04ad2c 怎么安装?

在客户端按下面的安装命令接入:docker run -i --rm ghcr.io/nulab/backlog-mcp-server node build/index.js --export-translations。装完在客户端配置里引用同一个包或端点。

Backlog MCP Server 1f04ad2c 怎么在 Claude Desktop、Claude、Cursor 里配置?

它的传输方式是stdio · 本地进程,在客户端的 MCP 配置里填好启动命令和参数,保存后完全退出客户端再打开,配置才会生效。

Backlog MCP Server 1f04ad2c 收费吗?

本站不代收费用,也没有标注具体价格。是否收费取决于服务本身和背后的接口额度,正式使用前按官方说明确认。

Backlog MCP Server 1f04ad2c 需要 API Key 吗?

资料标注需要会话登录认证,使用前准备好对应凭据,并按最小权限配置。

Backlog MCP Server 1f04ad2c 支持哪些客户端?

资料标注支持 Claude Desktop、Claude、Cursor、Cline;其他客户端只要能配置 MCP Server,也能用同样的命令或地址接入。

Backlog MCP Server 1f04ad2c 能提供哪些能力?

资料显示约 52 个工具;能力标签包括 问题跟踪、API代理、TypeScript、Claude、Git集成。具体工具清单以官方文档为准。

Backlog MCP Server 1f04ad2c 需要什么运行环境?

运行时为 Docker;属于本地进程,需要在本机装好运行环境。

Backlog MCP Server 1f04ad2c 是本地还是远程?

它是本地进程(stdio),由客户端在本机拉起。传输方式为stdio · 本地进程。

热门专题

按标签挑服务器
全部专题 →

服务器信息以作者仓库与官方文档为准;分类、传输方式与鉴权方式的中文说明为中文整理,最近核验 2026-05-17。