frappe-mcp-server
A secure, audit-logged MCP server for Frappe that exposes specific DocTypes and operations to AI agents with granular permissions and field-level control.
README
MCP Server for Frappe
A secure, audit-logged Model Context Protocol (MCP) server for Frappe. This app allows you to expose specific DocTypes and operations to MCP clients (like AI agents) with strict granular permissions, ensuring secure and controlled access to your Frappe data.
Features
- Secure Authentication: dedicated
MCP Useraccess via Token or API Key. - Granular Allowlist: Explicitly allow specific DocTypes and operations (Create, Read, Update, Delete).
- Field-Level Control: Restrict which fields can be read or written.
- Audit Logging: Every request is logged with latency, status, and payload summaries.
- Strict Validation: All filters and inputs are validated against the allowlist.
Installation
-
Get the app:
bench get-app https://github.com/mascor/frappe-mcp-server -
Install on your site:
bench --site <your-site> install-app mcp_server
Configuration
1. Setup MCP User & Settings
You can use the included setup script to quickly configure the server with a default user and token:
bench --site <your-site> execute mcp_server.setup.setup_mcp
This will:
- Create a user
mcp@example.com(if not exists). - Generate a secure
X-MCP-Tokenand save it in MCP Server Settings. - Enable the MCP Server.
Alternatively, go to MCP Server Settings in Desk and configure manually.
2. Configure Allowlist
By default, no access is allowed. You must explicitly allow DocTypes.
- Go to MCP Doctype Allowlist.
- Create a new record for the DocType you want to expose (e.g.,
ToDo). - Check the operations you want to allow (
Allow Read,Allow Create, etc.). - (Optional) restricting fields and filters in the child tables for tighter security.
Usage
The server exposes standard MCP tools via the API.
Authentication
The server supports API Key authentication (recommended for MCP clients).
Include the Authorization header in all requests.
Tools
mcp_ping
Health check and version info.
curl -X POST https://<yoursite>/api/method/mcp_server.api.ping \
-H "Authorization: token <api_key>:<api_secret>"
search_docs
Search for documents with filters.
{
"doctype": "ToDo",
"filters": {"status": "Open"}
}
create_doc
Create a new document.
{
"doctype": "ToDo",
"data": {
"description": "New Task from MCP"
}
}
update_doc
Update an existing document.
{
"doctype": "ToDo",
"name": "TODO-0001",
"data": {
"status": "Closed"
}
}
Audit Log
Check MCP Audit Log in Desk to see a history of all requests, including success/failure status, latency, and diffs.
License
MIT
Quick Connect: MCP Client
To connect to this server from an AI Client (like Claude Desktop or Cursor), use the official Python Client:
Follow the installation and setup instructions in the client repository to get started.
Frappe MCP Server - Test Report
Available Commands
| Command | Description | Parameters |
|---|---|---|
ping |
Check server connection | None |
search_docs |
Search documents | doctype (required), filters (optional), fields (optional) |
get_doc |
Get a specific document | doctype (required), name (required) |
create_doc |
Create a new document | doctype (required), data (required) |
update_doc |
Update an existing document | doctype (required), name (required), data (required) |
get_meta |
Get DocType metadata | doctype (required) |
delete_doc |
Delete a document | doctype (required), name (required) |
Test Results Summary
Section A: Standard DocType Operations
| Test | DocType | Operation | Status |
|---|---|---|---|
| A1 | Note | create_doc | ✅ PASS |
| A2 | Note | get_doc | ✅ PASS |
| A3 | Note | update_doc | ✅ PASS |
| A4 | Note | search_docs | ✅ PASS |
| A5 | Note | delete_doc | ✅ PASS |
| A6 | Event | create_doc | ✅ PASS |
| A7 | Event | get_meta | ✅ PASS |
| A8 | Event | delete_doc | ✅ PASS |
Section B: Filters and Fields
| Test | Description | Status |
|---|---|---|
| B1 | search_docs with filters | ✅ PASS |
| B2 | search_docs with fields | ✅ PASS |
| B3 | search_docs with filters + fields | ✅ PASS |
Section C: Edge Cases / Error Handling
| Test | Description | Expected | Status |
|---|---|---|---|
| C1 | get_doc non-existent document | 404 Not Found | ✅ PASS |
| C2 | create_doc missing required field | 417 Expectation Failed | ✅ PASS |
| C3 | update_doc non-existent document | 404 Not Found | ✅ PASS |
| C4 | delete_doc non-existent document | 404 Not Found | ✅ PASS |
| C5 | get_meta non-existent DocType | 403 Forbidden | ✅ PASS |
Section D: Permissions
| Test | Description | Status |
|---|---|---|
| D1 | search_docs on allowed DocType (User) | ✅ PASS |
| D2 | get_meta on non-allowed DocType (Role) | ✅ PASS (403) |
🎉 Final Result: 18/18 Tests Passed
Test Examples
1. ping
Check server connection status.
Response:
{
"status": "ok",
"site": "your-site.example.com",
"user": "mcp_user@example.com"
}
2. get_meta
Get the structure of a DocType.
Request:
doctype: "ToDo"
Response:
{
"doctype": "ToDo",
"fields": [
{
"fieldname": "status",
"label": "Status",
"fieldtype": "Select",
"options": "Open\\nClosed\\nCancelled",
"reqd": 0,
"default": "Open"
},
{
"fieldname": "priority",
"label": "Priority",
"fieldtype": "Select",
"options": "High\\nMedium\\nLow",
"reqd": 0,
"default": "Medium"
},
{
"fieldname": "date",
"label": "Due Date",
"fieldtype": "Date",
"reqd": 0,
"default": "Today"
},
{
"fieldname": "description",
"label": "Description",
"fieldtype": "Text Editor",
"reqd": 1
}
],
"issingle": 0,
"istable": 0
}
3. create_doc
Create a new document.
Request:
doctype: "ToDo"
data: {
"status": "Open",
"priority": "High",
"description": "Test task created via MCP"
}
Response:
{
"name": "abc123xyz",
"owner": "mcp_user@example.com",
"creation": "2026-01-21 10:18:41.295624",
"modified": "2026-01-21 10:18:41.295624",
"modified_by": "mcp_user@example.com",
"docstatus": 0,
"status": "Open",
"priority": "High",
"date": "2026-01-21",
"description": "Test task created via MCP",
"doctype": "ToDo"
}
4. search_docs
Search for documents with optional filters and fields.
Request (basic):
doctype: "ToDo"
Response:
[
{"name": "todo001"},
{"name": "todo002"},
{"name": "todo003"}
]
Request (with filters and fields):
doctype: "ToDo"
filters: {"status": "Open"}
fields: ["name", "status", "priority", "description"]
Response:
[
{
"name": "todo001",
"status": "Open",
"priority": "Medium",
"description": "Test task 1"
},
{
"name": "todo002",
"status": "Open",
"priority": "High",
"description": "Test task 2"
}
]
5. get_doc
Get a specific document by name.
Request:
doctype: "ToDo"
name: "abc123xyz"
Response:
{
"name": "abc123xyz",
"owner": "mcp_user@example.com",
"status": "Open",
"priority": "High",
"date": "2026-01-21",
"description": "Test task created via MCP",
"creation": "2026-01-21 10:18:41.295624",
"modified": "2026-01-21 10:18:41.295624",
"modified_by": "mcp_user@example.com",
"docstatus": 0
}
6. update_doc
Update an existing document.
Request:
doctype: "ToDo"
name: "abc123xyz"
data: {
"status": "Closed",
"priority": "Low"
}
Response:
{
"name": "abc123xyz",
"owner": "mcp_user@example.com",
"creation": "2026-01-21 10:18:41.295624",
"modified": "2026-01-21 10:18:59.096886",
"modified_by": "mcp_user@example.com",
"docstatus": 0,
"status": "Closed",
"priority": "Low",
"date": "2026-01-21",
"description": "Test task created via MCP",
"doctype": "ToDo"
}
7. delete_doc
Delete a document.
Request:
doctype: "ToDo"
name: "abc123xyz"
Response:
{
"status": "deleted",
"name": "abc123xyz"
}
Error Response (document not found):
HTTP 404 Not Found
Setup Requirements
-
Install the MCP Server app on your Frappe site
bench get-app https://github.com/[repo]/frappe-mcp-server bench --site [your-site] install-app mcp_server -
Create an API user (e.g.,
mcp_user@example.com) -
Generate API keys for the user
- Go to User → API Access → Generate Keys
-
Assign appropriate roles to the API user (e.g., System Manager)
-
Configure the Allowlist
- Go to "MCP Doctype Allowlist"
- Add DocTypes you want to expose via MCP
- Configure allowed operations (Read, Create, Update, Delete, Meta)
-
Configure the MCP client with the API credentials
Allowlist Configuration
The MCP Server uses an Allowlist to control which DocTypes are accessible. For each DocType you can configure:
| Option | Description |
|---|---|
| Allow Read (Get, Search) | Permits get_doc and search_docs |
| Allow Create | Permits create_doc |
| Allow Update | Permits update_doc |
| Allow Delete | Permits delete_doc |
| Allow Meta (Schema) | Permits get_meta |
| Max Page Length | Limits results for search_docs (0 = default) |
| Allowed Fields (Read) | Restrict which fields are returned (empty = all) |
| Allowed Filters | Restrict which fields can be used in filters (empty = all) |
Error Codes
| HTTP Code | Meaning | Common Causes |
|---|---|---|
200 |
Success | Operation completed |
403 |
Forbidden | DocType not in Allowlist or operation not permitted |
404 |
Not Found | Document or DocType does not exist |
417 |
Expectation Failed | Validation error (e.g., missing required field) |
500 |
Internal Server Error | Server-side error (check logs) |
Troubleshooting
| Issue | Cause | Solution |
|---|---|---|
403 Forbidden |
Missing permissions | Assign roles to API user or add @frappe.whitelist() decorator |
403 Forbidden on all calls |
Missing Allowlist entry | Add DocType to MCP Doctype Allowlist |
403 Forbidden on specific operation |
Operation not enabled | Enable the operation in Allowlist |
403 Forbidden with filters |
Filter field not allowed | Add field to Allowed Filters or leave empty for all |
404 Not Found |
Document doesn't exist | Verify document name |
417 Expectation Failed |
Module not found | Install/reinstall the MCP Server app |
417 Expectation Failed |
Validation error | Check required fields in DocType |
500 Internal Server Error |
Code error | Run bench --site [site] logs |
Security Notes
- Only DocTypes explicitly added to the Allowlist are accessible
- Each operation (Read, Create, Update, Delete, Meta) must be explicitly enabled
- API authentication is required for all operations
- Consider restricting sensitive DocTypes (User, Role, etc.) in production
Security Audit Report
Overview
The Frappe MCP Server has undergone comprehensive security testing to ensure it is safe for production use. All 15 security tests have passed successfully.
Security Test Results
Authentication & Authorization
| Test | Description | Result |
|---|---|---|
| S1 | Sensitive fields (password, api_key, api_secret) not accessible | ✅ PASS |
| S2 | System DocTypes (DocType) blocked | ✅ PASS |
| S3 | Role DocType blocked | ✅ PASS |
| S4 | System Settings blocked | ✅ PASS |
| S5 | Email Account (credentials) blocked | ✅ PASS |
| S6 | Allowlist self-modification blocked | ✅ PASS |
Injection Attacks
| Test | Description | Result |
|---|---|---|
| S7 | SQL Injection in filters | ✅ PASS |
| S8 | SQL Injection in document name | ✅ PASS |
| S12 | XSS in document content | ✅ PASS (sanitized) |
| S14 | Path Traversal attack | ✅ PASS |
Data Protection
| Test | Description | Result |
|---|---|---|
| S9 | Error Log access blocked | ✅ PASS |
| S10 | Scheduled Job Log access blocked | ✅ PASS |
| S11 | File DocType access blocked | ✅ PASS |
| S13 | Sensitive fields auto-filtered from User | ✅ PASS |
| S15 | Privilege escalation prevented | ✅ PASS |
Security Features
1. Allowlist-Based Access Control
Only DocTypes explicitly added to the MCP Doctype Allowlist are accessible via the API. Each DocType can be configured with granular permissions:
- Allow Read (Get, Search)
- Allow Create
- Allow Update
- Allow Delete
- Allow Meta (Schema)
2. Automatic Sensitive Field Filtering
The following fields are automatically removed from all API responses, regardless of user permissions or Allowlist configuration:
api_keyapi_secretpasswordnew_passwordreset_password_key
This server-side filter prevents credential leakage even if the User DocType is in the Allowlist.
3. Input Sanitization
- SQL Injection Protection: All user inputs are parameterized through Frappe's ORM
- XSS Protection: HTML content is automatically sanitized (e.g.,
<script>→<script>) - Path Traversal Protection: DocType names are validated against the Allowlist
4. Protected System DocTypes
The following sensitive DocTypes are blocked by default (not in Allowlist):
DocType- System schemaRole- Permission rolesSystem Settings- Global configurationEmail Account- Email credentialsError Log- System errors (may contain sensitive data)Scheduled Job Log- Background job logsFile- Uploaded filesMCP Doctype Allowlist- Prevents self-modification
Security Best Practices
For Production Deployment
- Minimize Allowlist: Only add DocTypes that are strictly necessary
- Restrict Operations: Enable only required operations (Read/Create/Update/Delete)
- Use Field Restrictions: Limit which fields can be read or used in filters
- Separate API User: Create a dedicated user for MCP with minimal roles
- Monitor Access: Regularly review API access logs
- Rotate API Keys: Periodically regenerate API credentials
Allowlist Configuration Example
DocType: Customer
├── Allow Read: ✅
├── Allow Create: ❌
├── Allow Update: ❌
├── Allow Delete: ❌
├── Allow Meta: ✅
├── Allowed Fields: ["name", "customer_name", "email"]
└── Allowed Filters: ["name", "customer_name"]
This configuration allows read-only access to Customer records with restricted fields.
Tested Attack Vectors
SQL Injection
# Attempted attack
search_docs(doctype="ToDo", filters={"status": "Open'; DROP TABLE tabToDo; --"})
# Result: Query safely parameterized, no data loss
XSS (Cross-Site Scripting)
# Attempted attack
create_doc(doctype="ToDo", data={"description": "<script>alert('XSS')</script>"})
# Result: Content sanitized to "<script>alert('XSS')</script>"
Path Traversal
# Attempted attack
get_doc(doctype="../../../etc/passwd", name="root")
# Result: 403 Forbidden (DocType not in Allowlist)
Privilege Escalation
# Attempted attack
update_doc(doctype="User", name="attacker@example.com", data={"user_type": "Administrator"})
# Result: 403 Forbidden (Update not allowed on User DocType)
Compliance Summary
| Category | Status |
|---|---|
| Authentication | ✅ API Key/Secret required |
| Authorization | ✅ Allowlist-based access control |
| Data Protection | ✅ Sensitive fields auto-filtered |
| Input Validation | ✅ SQL/XSS/Path Traversal protected |
| Audit Trail | ✅ Standard Frappe logging |
Final Result
╔══════════════════════════════════════════════════════════════╗
║ SECURITY AUDIT PASSED ║
║ ║
║ 15/15 Tests Successful ║
║ ║
║ ✅ Ready for Production Deployment ║
╚══════════════════════════════════════════════════════════════╝
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。