BitrixMCP
Universal MCP server for the Bitrix24 REST API, enabling full read and write access to CRM, tasks, calendar, disk, and more. Supports any MCP client with stdio or Streamable HTTP transport.
README
bitrix-mcp
English · Русский
Universal, full-featured, portable MCP server for the Bitrix24 REST API. Read and write. Not tied to any one application — it's a generic Bitrix24 gateway that any MCP client or agent can mount (Claude Code, Claude Desktop, Cursor, Windsurf, Cline, or your own Python/Node agent).
- Language: Python + FastMCP (official MCP SDK)
- Transports:
stdio(default, most portable/reliable) and Streamable HTTP (stateless JSON — no fragile long-lived SSE bridge) - Coverage: universal
b24_call/b24_batchreach 100% of the REST API; 88 typed tools cover the high-traffic domains with the tricky bits handled.
Why this exists / what it fixes
Rebuilt from field notes on a previous wrapper. The bugs that motivated it are fixed by design, not patched around:
| Old behavior | Fix here |
|---|---|
filter silently ignored (groups_list, users_list), full-portal dumps → timeouts |
Params sent as JSON POST body, so nested filter/select/order are parsed correctly by Bitrix. Real pagination with a page cap. |
Access errors swallowed into a fake "0 results" (read_pipelines etc.) |
Errors are never swallowed — a Bitrix error/error_description always surfaces with its code (e.g. ACCESS_DENIED). |
calendar_list returned 0 without explicit ownerId |
owner_id auto-resolves to the acting user. |
| Scrum kanban read from the wrong place | Correct flow baked in: active-sprint filter + tasks.api.scrum.kanban.getStages (b24_scrum_board does it in one call). |
Fragile mcp-remote SSE session drops / hangs |
Prefer stdio (no bridge) or stateless Streamable HTTP. |
department.get has no server-side filter at all (a Bitrix API limitation, undocumented) — any filter was silently ignored and the whole department tree (95+ rows) came back regardless |
b24_department_get filters client-side after a full fetch, so filter/ID genuinely narrow the result instead of quietly dumping everything. |
Bitrix sometimes reports a failure as {"error": "", "error_description": "Access denied."} — an empty-string error code — which a naive truthiness check (if data.get("error")) misses, losing the code and message to a generic HTTP-status fallback |
Checked by key presence, not truthiness — code/message always reflect what Bitrix actually said. |
calendar.event.add / .update silently drop attendees unless is_meeting is also set — 200 OK, event created, nobody invited, no error anywhere |
is_meeting is auto-set to 'Y' whenever attendees is non-empty and not already specified. |
Moving a task on a Scrum sprint board via tasks.task.update's STAGE_ID is accepted with no error and reads back correctly, but the card does not actually move on the real board (confirmed live: reload the page, still in the old column) |
New b24_scrum_task_move calls the board-aware tasks.api.scrum.kanban.addTask instead — the one that actually relocates the card. b24_task_update/b24_tasks_list now document this trap rather than silently mis-teaching it. |
Install
uv sync # create venv + install
# or, as a tool on PATH:
uv tool install . # exposes the `bitrix-mcp` command
Configure
Set the default webhook (see .env.example):
export BITRIX_WEBHOOK_URL="https://your-portal.bitrix24.ru/rest/1/xxxxxxxx/"
# optional:
export BITRIX_READ_ONLY=1 # block all writes
The webhook comes from Bitrix: Profile → Webhooks → inbound webhook, format
https://<portal>/rest/<user_id>/<token>/. The token is a credential — keep it
out of source control (.env is gitignored).
Auth precedence per call: personal_webhook → webhook_url →
X-B24-Webhook HTTP header → BITRIX_WEBHOOK_URL. Pass personal_webhook to
act (and write) as a specific user.
Run
bitrix-mcp # stdio (default)
bitrix-mcp --http # Streamable HTTP on 127.0.0.1:8000/mcp
bitrix-mcp --http --host 0.0.0.0 --port 5015 # shared network service
Connect a client
Claude Code (stdio, recommended):
claude mcp add -s user bitrix24 -- uv run --directory C:/Scripts/BitrixMCP bitrix-mcp
Repo-shared .mcp.json (stdio):
{
"mcpServers": {
"bitrix24": {
"command": "uv",
"args": ["run", "--directory", "C:/Scripts/BitrixMCP", "bitrix-mcp"],
"env": { "BITRIX_WEBHOOK_URL": "https://your-portal.bitrix24.ru/rest/1/xxxx/" }
}
}
}
Claude Code (HTTP):
bitrix-mcp --http --port 5015 # then, on the client:
claude mcp add -s user --transport http bitrix24 http://HOST:5015/mcp
Claude Desktop (stdio) — %APPDATA%\Claude\claude_desktop_config.json:
{
"mcpServers": {
"bitrix24": {
"command": "uv",
"args": ["run", "--directory", "C:/Scripts/BitrixMCP", "bitrix-mcp"],
"env": { "BITRIX_WEBHOOK_URL": "https://your-portal.bitrix24.ru/rest/1/xxxx/" }
}
}
}
(For a remote HTTP instance, Desktop still needs the mcp-remote bridge; stdio
above avoids it entirely.)
Tool catalog (88)
Universal — b24_call, b24_batch, b24_test_connection, b24_list_methods
CRM — b24_crm_list, b24_crm_get, b24_crm_fields, b24_crm_add, b24_crm_update, b24_crm_delete, b24_crm_timeline_comment_add, b24_crm_timeline_comment_list, b24_crm_category_list (pipelines), b24_crm_status_list (stages/dictionaries), b24_crm_activity_list, b24_crm_activity_add, b24_crm_activity_delete, b24_crm_productrows_get, b24_crm_productrows_set, b24_crm_currency_list, b24_crm_requisite_list, b24_crm_deal_contacts_get, b24_crm_deal_contacts_set (classic entities and SPA via entity_type_id)
Tasks — b24_tasks_list, b24_task_get, b24_task_add, b24_task_update, b24_task_complete, b24_task_delete, b24_task_comments_list, b24_task_comment_add, b24_task_stages_get, b24_task_checklist_list, b24_task_checklist_add, b24_task_elapsed_add, b24_task_result_list
Scrum — b24_scrum_sprint_list, b24_scrum_kanban_stages, b24_scrum_board, b24_scrum_task_move
Calendar — b24_calendar_event_list, b24_calendar_section_list, b24_calendar_event_add, b24_calendar_event_update, b24_calendar_event_delete
Disk — b24_disk_storage_list, b24_disk_folder_items, b24_disk_file_get, b24_disk_file_content (server-side download → base64), b24_disk_folder_add, b24_disk_file_upload, b24_disk_file_delete
Users/structure — b24_user_get, b24_user_search, b24_user_current, b24_department_get
Groups (workgroups) — b24_group_list, b24_group_users, b24_group_create, b24_group_update, b24_group_delete
Messaging — b24_im_recent, b24_im_dialog_messages, b24_im_message_add, b24_im_notify_personal, b24_im_user_get, b24_im_chat_create, b24_im_chat_user_add, b24_feed_post_add
Lists (universal lists) — b24_lists_get, b24_lists_element_list, b24_lists_element_add, b24_lists_element_update, b24_lists_element_delete
Catalog / products — b24_catalog_list, b24_catalog_section_list, b24_catalog_product_list, b24_catalog_product_get, b24_catalog_product_add, b24_catalog_product_update, b24_crm_product_list
Sale (orders) — b24_sale_order_list, b24_sale_order_get
Documents — b24_documentgenerator_templates, b24_documentgenerator_add
Bizproc — b24_bizproc_template_list, b24_bizproc_start
Telephony — b24_telephony_statistics
Anything still not typed here is reachable through b24_call (e.g. mail,
open-lines, sale basket writes, admin/app-placement methods).
Retrospective-app integration
This server has no knowledge of any downstream app. An agent connects to both
this server and your app's MCP, reads Bitrix here, and relays into the app's
contract. Field names from b24_tasks_list / b24_calendar_event_list map
directly onto PushSprintTask / PushCalendarEvent, so the mapping is trivial —
but that translation lives in the agent, not here.
Documentation
- docs/USAGE.md — how to drive the tools: auth, filters, pagination, batch, errors, recipes.
- ARCHITECTURE.md — components, call flow, deployment; diagrams in docs/diagrams/.
- ROADMAP.md — status, history, and out-of-scope boundaries.
Development
uv sync # install runtime + dev deps
uv run pytest -q # offline unit tests (no portal needed)
uv run python scripts/smoke.py "<webhook>" # live read-only access map (run from a network with portal access)
Diagrams are regenerated with java -jar plantuml.jar -tpng docs/diagrams/*.puml.
Notes on limits
fetch_all=trueis capped byBITRIX_MAX_PAGES(default 40 pages ≈ 2000 records) and reportstruncated: truewhen it hits the cap — it never silently stops short.- The read-only guard classifies writes by method verb; typed write tools are
always classified correctly.
b24_call/b24_batchuse the heuristic.
Support author
<p align="center"> <img src="donate-qr.png" alt="Donate QR" width="200"/> </p>
BTC: bc1q3frrup5neh7nhfg944etu2agd4j9u0vg3jyee6
ETH(Arbitrum): 0x43B349d8Cea83215D707EBa3bc35e9917f746b0a
TRX: THSzvy49KNeqRjXsGkurh2A5G4avV4RgN4
XRP: rLWZjS3DMupC4ZdXCX3BVYn4dEtC3iNhgy
SOL: 3xwfybxJ6Tz5t6pjBBkL5yYQCZo6wfbv932UNA4ThdP8
ADA: addr1q926ys75jp5wn2pv32a3t8r8pdhr7w02v0t9j4a8pmg0ruww5rlkctu4lnz2hfcwa5qfn3zhsd0s23r22uqwzx9gu6cq5c4e76
TON: UQC4qlAOD9Nly4K_66GJ_yCsSM3x2sB0vZ2GrBQbc--gZUui
DOGE: DTjNYmbtymzcjUiV4MsZY8MP4dM7MJ6qLC
XMR: 44qRqM6YtnxXUhkgCFqDDrKMPjWriu69FLBoop8Kwp7e1VQsBUJoVQ8JYQjfMV5C6uidTUgSSyoJ65mq8aYG2esZ1rrqfwt
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。