mcp server
grist-go-mcp
MCP Server for Grist written in Go
Set up
Runs on your computer. Pick your agent and operating system.
This project does not publish an install method we can turn into a config. Follow the README instead.
README
synced Oct 23, 2026Grist MCP Server
[⚠️] Work in progress, not ready for production yet.
[⚠️] TODO: Must minimize prompts and responses to limit tokens usage (around 8000 tokens for the sample request)
[⚠️] Disclaimer: This is just a for-fun project, not actively maintained, but happy to share and contributions are welcome !
Functions:
- ✅
list_orgs: List all organizations - ✅
create_workspace: Creates a new workspace in organization - ✅
create_new_docs: Creates one or more documents in workspace - ✅
create_new_tables: Add new tables with columns to a document - ☑️
query_sql: Query SQL (API not yet available) - ☑️
add_record: Add a record in a document (API not yet available) - ☑️
get_record: Get a record in a document (API not yet available)
Examples
System prompt example available at docs/SYSTEM_PROMPT.md
Prompt:
Create a personal finance tracker in Grist with the following structure:
- Tables needed:
- Transactions: columns for Date, Description, Category, Amount, Payment Method, and Notes
- Categories: columns for Category Name, Type (Income/Expense), Budget Limit
- Monthly Summary: columns for Month, Total Income, Total Expenses, Net Balance
Getting started
- Clone the project
$ git clone https://github.com/quentinchampenois/grist-mcp.git
$ cd grist-mcp
- Build the server
$ go mod tidy
$ go build -o grist-mcp-server
💡 Binary grist-mcp-server must be present in $PATH to be callable
🚀 MCP Server is ready to use ! But how ?
If you try to run it directly $ ./grist-mcp-server process will run but nothing happens, and it's the expected behaviour.
Use @modelcontextprotocol/inspector to debug
Package @modelcontextprotocol/inspector starts a WebUI to test directly your MCP server, test output based on provided input.
Requirements
- Node
How to
Start inspector as following
$ npx @modelcontextprotocol/inspector grist-mcp-server
Access WebUI at http://localhost:6274/?MCP_PROXY_AUTH_TOKEN=<GENERATED_TOKEN>
Configure client using Cherry studio and Scaleway AI provider
In this example we will use Cherry Studio as client and model qwen3-235b-a22b-instruct-2507 from Scaleway provider.
Requirements
- Configure account on Scaleway and generate API token (~5 minutes)
- Cherry studio installed
Configure Qwen3-235b from Scaleway on Cherry Studio
Open Cherry Studio and configure the Scaleway provider: (cf: README on another project for full chery studio steps) 💡 Very important, setup the environment variables in your cherry studio MCP configuration :
GRIST_ENDPOINT=http://localhost:8484GRIST_API_KEY=<YOUR_API_KEY>, API_KEY from GRIST WebUI
🚀 Once Cherry studio let's give a try with a basic prompt
I want to manage my tennis club within Grist, including: members, available fields and hours of lessons.
You must create in Grist :
* 1 new workspace in my Grist organization
* 3 documents in this newly created workspace
Example:
Prompt:
Result:
Extra
- Source Cherry studio: https://github.com/CherryHQ/cherry-studio
- Source Scaleway AI (Generative API): https://www.scaleway.com/en/docs/generative-apis/quickstart/
- Go MCP SDK: https://github.com/modelcontextprotocol/go-sdk/tree/main
- MCP Inspector: https://modelcontextprotocol.io/docs/tools/inspector