Kobo MCP Server
Enables reading and writing to KoBoToolbox via MCP, including creating forms, deploying them, attaching media, and submitting data, all from chat.
README
Kobo MCP Server
An MCP server that lets Claude (Desktop, Code, or any MCP client) read and write to KoBoToolbox. Create forms from a chat, deploy them, attach media, submit data, all without leaving the conversation.
Forked and extended from PYI-SOE-LTD/kobo-mcp-server, which was read-only. This fork adds 10 new write tools (form creation in JSON or XLSForm, multilingual support, media attachments, deploy, update, clone, submissions).
✨ Features
- 🌍 Multilingual forms — create surveys in multiple languages at once (e.g. French + Lingala)
- 📝 Two creation modes — JSON schema (simple) or XLSForm Excel (standard)
- 🚀 Full lifecycle — create → attach media → deploy → submit → update → redeploy → clone
- 🖼️ Media attachments — logos, reference images, audio prompts, video, from local file or URL
- 🔒 Token in
.env— never hardcoded, never committed
🧰 16 tools exposed
Read (6)
listProjects · getProjectSummary · getProjectSchema · queryProjectData · getCountsBy · getTimeWindowSummary
Write — forms (7)
createForm · createFormFromXLSForm · deployForm · redeployForm · updateForm · createSubmission · cloneForm
Write — media (3)
attachMedia · listMedia · removeMedia
🚀 Quick start
1. Get a KoBo API token
- Global: https://kf.kobotoolbox.org/token/
- OCHA/Humanitarian: https://eu.kobotoolbox.org/token/
- Self-hosted:
https://<your-host>/token/
2. Install
git clone https://github.com/tagnonkoua-ux/kobo-mcp-server.git
cd kobo-mcp-server
pip install -r requirements.txt
3. Configure
cp .env.example .env
# Edit .env and paste your KOBO_TOKEN
4. Register the MCP server in Claude
Claude Code (recommended):
claude mcp add kobo python /absolute/path/to/kobo-mcp-server/mcp_server.py
Or edit ~/.claude.json manually:
{
"mcpServers": {
"kobo": {
"command": "python",
"args": ["/absolute/path/to/kobo-mcp-server/mcp_server.py"],
"env": {
"KOBO_BASE": "https://kf.kobotoolbox.org/api/v2",
"KOBO_TOKEN": "your_token_here",
"PYTHONIOENCODING": "utf-8"
}
}
}
}
Claude Desktop: copy the same block into claude_desktop_config.json (see claude_desktop_config.json for the template).
💡 On Windows, use double-backslashes in JSON paths:
"C:\\Users\\you\\kobo-mcp-server\\mcp_server.py".
5. Restart Claude
The 16 Kobo tools should now appear. Try:
"Create a customer satisfaction survey with 5 questions in French and English, deploy it, and give me the public link."
📐 Simple JSON schema for createForm
Mono-lingual
{
"title": "Customer satisfaction",
"language": "English (en)",
"questions": [
{"name": "your_name", "type": "text", "label": "Your name?", "required": true},
{"name": "rating", "type": "integer", "label": "Rate us 1-10",
"constraint": ". >= 1 and . <= 10"},
{"name": "city", "type": "select_one", "label": "City?",
"choices": ["Brazzaville", "Pointe-Noire", "Dolisie"]}
]
}
Multilingual (label as dict)
{
"title": "Survey",
"languages": ["English (en)", "Français (fr)", "Lingala (ln)"],
"default_language": "Français (fr)",
"questions": [
{"name": "name", "type": "text", "required": true,
"label": {"en": "Your name?", "fr": "Votre nom ?", "ln": "Nkombo na yo ?"}},
{"name": "city", "type": "select_one",
"label": {"en": "City?", "fr": "Ville ?", "ln": "Engumba ?"},
"choices": [
{"name": "bzv", "label": {"en": "Brazzaville", "fr": "Brazzaville", "ln": "Brazzaville"}},
{"name": "pnr", "label": {"en": "Pointe-Noire", "fr": "Pointe-Noire", "ln": "Pointe-Noire"}}
]}
]
}
Supported types
text, integer, decimal, select_one, select_multiple, date, time, datetime, note, geopoint, geotrace, geoshape, image, audio, video, file, barcode, calculate, acknowledge, range, rank, begin_group/end_group, begin_repeat/end_repeat.
Per-question fields
required · relevant (skip logic) · constraint · constraint_message · default · appearance · calculation · read_only · parameters · hint.
🧪 Testing
Run the live test suite against your Kobo account (creates ephemeral forms, cleans them up automatically):
python tests/test_write_ops.py
Expected: 11/11 passed.
The tests cover: list, create (mono & multilingual), deploy, submit, update + redeploy, clone, XLSForm upload, media attach/list/remove.
🔒 Security
- Token lives in
.envonly —.envis.gitignored. - The original upstream repo had a hardcoded token in
run.sh. This fork loads from.envinstead. - The server logs only the first 6 chars of the token to stderr (e.g.
abc123...). - All HTTP goes to your configured Kobo host (default
kf.kobotoolbox.org) and KoBoCAT (kc.kobotoolbox.org) for submissions.
If you accidentally leak your token, regenerate it at https://kf.kobotoolbox.org/token/ — the old one stops working instantly.
🌐 Other Kobo instances
Just change KOBO_BASE:
# OCHA / humanitarian
KOBO_BASE=https://eu.kobotoolbox.org/api/v2
# Self-hosted
KOBO_BASE=https://kobo.your-org.example.com/api/v2
The submission code derives the KoBoCAT host automatically by replacing kf. / kpi. with kc. in the hostname.
🛠️ Architecture
kobo-mcp-server/
├── kobo_client.py # HTTP client for the KPI v2 + KoBoCAT APIs
├── xlsform_builder.py # JSON → .xlsx XLSForm converter
├── mcp_server.py # MCP stdio server (16 tools)
├── server.py # Optional FastAPI HTTP server (legacy)
├── tests/
│ └── test_write_ops.py
├── requirements.txt
├── .env.example
└── README.md
📜 License
MIT — see LICENSE. Original work © PYI-SOE-LTD. Extensions © 2026.
🙏 Credits
- Base server: PYI-SOE-LTD/kobo-mcp-server
- KoBoToolbox API: kobotoolbox/kpi
- MCP protocol: modelcontextprotocol.io
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。