lark-hermes-mcp
An MCP server that exposes Feishu/Lark capabilities as function tools to MCP-compatible agents like Hermes and Claude Desktop. It provides 57 tools for messaging, document management, calendar operations, and task management through both hand-written fallback tools and bridged OpenClaw SDK functionality.
README
lark-hermes-mcp
Thin MCP stdio server that exposes Feishu (飞书) / Lark capabilities as function tools to Hermes, Claude Desktop, or any other MCP-compatible agent.
- 17 fallback tools (hand-written) for messaging, bitable, calendar, docs, and task operations
- 36 tools bridged from
@larksuite/openclaw-larkvia a shim adapter - 4 OAuth tools (
lark_oauth_start/lark_oauth_complete/lark_oauth_status/lark_oauth_revoke) driving OpenClaw's Device Flow for user-access-token authorization
Transport: stdio (stdout is JSON-RPC only, pino logs go to stderr) Auth: tenant_access_token + user-access-token (OAuth Device Flow) SDK:
@larksuiteoapi/node-sdk+@larksuite/openclaw-lark
Requirements
- Node.js ≥ 22
- A Feishu or Lark custom app (自建应用) that you own
- An MCP-compatible client (e.g. Hermes, Claude Desktop, or any stdio MCP host)
Installation
Step 1 — Register your own Feishu / Lark app
This project does not ship with any pre-registered app credentials. Every user must create their own app on the Feishu/Lark open platform.
- Go to https://open.feishu.cn/app (国内 Feishu) or https://open.larksuite.com/app (海外 Lark).
- Click "创建企业自建应用" / "Create Custom App". Give it a name and icon.
- After creation, open the app dashboard:
- 凭证与基础信息 / Credentials & Basic Info page → copy your
App ID(formatcli_xxxxxxxxxxxxxxxx) andApp Secret. - 权限管理 / Permissions & Scopes page → enable the scopes you plan to
use. For the tools bundled here, the core set is:
im:message,im:chat(messaging)bitable:app(多维表格)docx:document,drive:drive(docs)calendar:calendar(日历)- Task-related scopes if you want task tools
- If you plan to use the OAuth (user-access-token) tools, also enable
Device Flow / OAuth in your app settings. The OAuth tools request
granular per-API scopes at runtime —
lark_oauth_startwill print the exact scope list on first call so you know what to enable.
- 凭证与基础信息 / Credentials & Basic Info page → copy your
- 发布 / Publish the app version in the app dashboard so your scopes take effect.
Step 2 — Clone and install
git clone https://github.com/WilliamMo101/lark-hermes-mcp.git
cd lark-hermes-mcp
npm install # triggers postinstall patches (see "Upstream & Patches" below)
npm run build
Step 3 — Configure your credentials
Standalone (any MCP host):
cp .env.example .env
# then edit .env and paste the App ID / App Secret you got in Step 1
.env is git-ignored and will never be committed.
Under Hermes: you do not need a .env file. Put the variables directly in
profiles/<your-agent>/config.yaml under mcp_servers.lark.env:
mcp_servers:
lark:
command: /path/to/node
args:
- /path/to/lark-hermes-mcp/dist/server.js
env:
LARK_APP_ID: "cli_xxxxxxxxxxxxxxxx"
LARK_APP_SECRET: "xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"
LARK_DOMAIN: "Feishu"
LARK_ENABLED_TOOLSETS: "messaging,docs,bitable,calendar,other"
LARK_LOG_LEVEL: "info"
timeout: 120
connect_timeout: 30
tools:
resources: false
prompts: false
LARK_DOMAIN: mainland Feishu →Feishu; overseas Lark →Lark. The wrong value will be rejected by the OAuth service (invalid_client).
LARK_ENABLED_TOOLSETSmust includeotherif you want the OAuth tools (lark_oauth_*) and task tools to be exposed.
Step 4 — Run and verify
node dist/server.js # starts the stdio MCP server
Or hook it up to your MCP host and invoke feishu_get_user or
lark_oauth_status as a smoke test.
Hermes exposes the tools as mcp_lark_<name> (underscores, per
mcp_tool.py:sanitize_mcp_name_component).
Example paths in this README (such as
/root/.hermes/mcp-servers/…in older snippets) reflect the author's own Hermes-on-WSL layout. Adapt them to wherever you check the repo out.
Fallback tools (17)
These are hand-written specs in src/adapter/fallback.ts:
| toolset | name | what |
|---|---|---|
| messaging | sendMessageFeishu |
send IM to chat / user / email |
| messaging | sendCardFeishu |
send interactive card |
| messaging | replyMessageFeishu |
reply to a message_id |
| messaging | listMessagesFeishu |
list recent messages in a chat |
| bitable | bitableListRecords |
paged record list with filter/sort |
| bitable | bitableCreateRecord |
insert record |
| bitable | bitableUpdateRecord |
update record |
| calendar | calendarListCalendars |
list calendars |
| calendar | calendarCreateEvent |
create event |
| calendar | calendarListEvents |
list events in range |
| docs | docxGetRawContent |
raw-text fetch of a docx |
| docs | docxListBlocks |
block tree of a docx |
| other | selfCheck |
diagnostics: credentials + token acquisition |
| other | feishu_get_user |
get current user info |
| other | task-related helpers | (see fallback.ts) |
OpenClaw-bridged tools (36)
src/adapter/shim.ts calls registerXxxTools(api) against
@larksuite/openclaw-lark's internal registrations, wraps each tool with
typebox → JSON Schema flattening (for OpenAI function-calling compatibility),
and injects withTicket context via AsyncLocalStorage.
Tools appear with the feishu_ / mcp_doc_ prefix, e.g.
feishu_bitable_app, feishu_calendar_event, feishu_im_chat_messages.
OAuth tools (4)
lark_oauth_start— begin Device Flow (prints user_code + verification URL)lark_oauth_complete— poll for token after user authorizes in browserlark_oauth_status— check stored user token status (valid / needs_refresh / expired)lark_oauth_revoke— revoke stored user token
Tokens are encrypted (AES-256-GCM) and stored under
~/.local/share/openclaw-feishu-uat/.
Smoke test (standalone, without an MCP host)
export LARK_APP_ID=cli_xxx LARK_APP_SECRET=xxx LARK_DOMAIN=Feishu
(
echo '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-03-26","capabilities":{},"clientInfo":{"name":"smoke","version":"0"}}}'
echo '{"jsonrpc":"2.0","method":"notifications/initialized","params":{}}'
echo '{"jsonrpc":"2.0","id":2,"method":"tools/list","params":{}}'
echo '{"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"selfCheck","arguments":{}}}'
) | node dist/server.js 2>/tmp/lark-mcp.err > /tmp/lark-mcp.out
jq '.result.tools | length' < /tmp/lark-mcp.out
jq 'select(.id==3)' < /tmp/lark-mcp.out
There is also a registration-count smoke test for the shim:
LARK_APP_ID=cli_xxx node scripts/shim-smoke.mjs
Upstream & Patches
This project builds on top of
@larksuite/openclaw-lark
(MIT License) — 36 of the exposed tools are bridged directly from its native
tool registrations via src/adapter/shim.ts.
scripts/postinstall-patches.mjs runs automatically after npm install and
applies idempotent patches inside node_modules/@larksuite/openclaw-lark/
so that the package loads cleanly under Node 22 CJS:
- Strip
import.meta.urlsyntax fromversion.js(replaced withcreateRequire-based resolution). - Same treatment for
token-store.js. - Build a minimal stub for
@openclaw/plugin-sdk(only the pieces the registrations actually import). - Widen the package's
exportsmap so deep./src/*imports resolve.
These patches only modify files inside your local node_modules/. The
upstream source is not modified, and the patches re-run safely on every
install.
Troubleshooting
- Nothing in the host's tool list — grep the host's log for
MCP server 'lark'. Common causes:LARK_APP_ID/LARK_APP_SECRETnot passed through — shell exports don't propagate if the host uses an allow-list env. Declare them in the MCP config block instead.dist/server.jsmissing — runnpm run build.- Wrong
nodebinary — must be Node ≥ 22.
- OAuth tools missing from tool list —
LARK_ENABLED_TOOLSETSmust includeother. invalid_scopeonlark_oauth_start— one of the requested scopes isn't granted yet on your app. Open the app's Permissions page, enable the scopes listed in the error, publish a new version.invalid_clienton OAuth —LARK_DOMAINis wrong for your app region (国内 =Feishu, 海外 =Lark).- stdout pollution (MCP client disconnects immediately) — some third-party lib wrote to stdout. Check stderr logs.
server.tsalready hijacksconsole.*and pino writes to stderr. - Rate limited client-side — tune
LARK_THROTTLE_BITABLE_RPSetc. via env.
Project layout
src/
server.ts # MCP entry, console hijack, handlers
auth.ts # @larksuiteoapi/node-sdk Client factory
log.ts # pino → stderr
toolsets.ts # toolset enum + env filter
util/throttle.ts # per-toolset token bucket
adapter/
index.ts # tool loader + toolset filter
fallback.ts # 17 hand-written fallback tool specs
shim.ts # OpenClaw bridge + schema flattening
oauth-tools.ts # 4 OAuth tools (Device Flow)
scripts/
postinstall-patches.mjs # idempotent node_modules patches
shim-smoke.mjs # registration-count smoke test
Credits
@larksuite/openclaw-larkby OpenClaw — MIT License. This project wraps and depends on it.- Model Context Protocol SDK by Anthropic.
- Feishu / Lark Open Platform APIs.
Disclaimer
This is a personal hobby project, provided as-is without any warranty. It is not an official product of Feishu, Lark, ByteDance, or OpenClaw.
这是一个个人兴趣项目,按"现状"提供,不附带任何担保。本项目与飞书/Lark、 字节跳动、OpenClaw 官方团队无隶属关系,出现问题请通过 GitHub Issues 反馈。
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.