asana-mcp
MCP server for the Asana API that allows AI agents to read and optionally write to Asana tasks, comments, and custom fields, with tiered access controls and no delete tools.
README
asana-mcp
MCP server + typed TypeScript client for the Asana API.
Two ways to use it:
- MCP server (
asana-mcpbin) — gives MCP clients (Claude Desktop, Claude Code, MCP Inspector, anymcporter-style runner) a guarded Asana tool surface over stdio. - Client library (
asana-mcp/client) — a small typed Asana REST client for direct in-process use in Node services, with retry/backoff and normalized errors.
Unofficial. Not affiliated with or endorsed by Asana, Inc. If you want Asana's hosted, OAuth-based MCP integration for interactive AI assistants, see Asana's official MCP server — this package exists for headless / personal-access-token use cases (service bots, automation, server-side integrations), which the official server does not support.
Design choices
- No delete tools, in any mode. Destructive removal is excluded from the tool surface by design.
- Write access is opt-in and tiered (
read_only→restricted→full), so an AI agent can be given exactly as much write capability as you intend — down to an allowlist of specific custom fields. - Zero baked-in account specifics. Workspace, fields, and tokens all arrive via environment variables.
MCP server
ASANA_ACCESS_TOKEN=0/123abc... npx asana-mcp
Environment
| Variable | Required | Description |
|---|---|---|
ASANA_ACCESS_TOKEN |
yes | Asana personal access token |
ASANA_MCP_WRITE_MODE |
no | read_only (default), restricted, or full |
ASANA_MCP_WRITABLE_CUSTOM_FIELDS |
no | Comma-separated custom field GIDs writable in restricted mode |
ASANA_MCP_DEFAULT_WORKSPACE |
no | Default workspace GID for list/search tools |
Write modes
| Mode | Registered write tools | Notes |
|---|---|---|
read_only |
none | Default. Read tools only. |
restricted |
asana_add_comment, asana_update_task |
asana_update_task may modify only custom fields whose GIDs are in ASANA_MCP_WRITABLE_CUSTOM_FIELDS; every other task field is rejected. |
full |
+ asana_create_task, unrestricted asana_update_task |
Still no delete tools. |
Tools
Read (always): asana_get_task, asana_list_tasks, asana_list_task_comments, asana_list_task_attachments, asana_get_user, asana_list_project_sections, asana_list_tasks_in_section, asana_typeahead_search.
Write (per mode, above): asana_add_comment, asana_update_task, asana_create_task.
Example MCP client config
{
"mcpServers": {
"asana": {
"command": "npx",
"args": ["asana-mcp"],
"env": {
"ASANA_ACCESS_TOKEN": "0/123abc...",
"ASANA_MCP_WRITE_MODE": "restricted",
"ASANA_MCP_WRITABLE_CUSTOM_FIELDS": "1200000000000001,1200000000000002",
"ASANA_MCP_DEFAULT_WORKSPACE": "1100000000000001"
}
}
}
}
Client library
import { AsanaClient, AsanaApiError } from 'asana-mcp/client';
const asana = new AsanaClient({ accessToken: process.env.ASANA_ACCESS_TOKEN! });
const task = await asana.getTask('1300000000000001', {
optFields: 'name,notes,assignee.name,custom_fields.display_value',
});
await asana.addComment(task.gid, {
text: 'Done!',
htmlText: '<body><strong>Done!</strong></body>', // tried first, falls back to text
});
await asana.setAssignee(task.gid, null); // unassign
for await (const t of asana.iterateTasks({
project: '1400000000000001',
optFields: 'name,completed',
limit: 100,
})) {
// auto-follows pagination
}
Client behavior
- Retries 429 (honoring
Retry-After), 500/502/503/504, and network errors with exponential backoff + jitter (default 3 retries, base 500 ms). - Per-attempt timeout (default 30 s) via
AbortController. - Errors throw
AsanaApiErrorwith.status,.body, and.request— non-retryable 4xx throws immediately. - Injectable
fetchImplfor tests; everything is constructor-configurable (baseUrl,maxRetries,timeoutMs,retryBaseDelayMs).
Surface: getTask, listTasks, iterateTasks, createTask, updateTask, setAssignee, getTaskStories, getStory, addComment, getTaskAttachments, getUser, getProjectSections, listTasksInSection, typeaheadSearch. (No delete methods — see design choices.)
Development
npm install
npm test # vitest unit suite (mocked fetch — no network, no token needed)
npm run build # tsc → build/
npm run inspector # poke the server with MCP Inspector
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。