mcp server best practices.md

For the complete documentation index, see llms.txt. Markdown versions of documentation pages are available by appending .md to page URLs; this page is available as Markdown.

MCP Server Best Practices

📌 Glue AI prompting tips for MCP

Mention the name of the MCP server you want to take action with
Use clear, very specific instructions
- e.g. "Update the status of Issue ID 12345 in Linear to 'In Progress'."
- Note: You may need ID numbers in order to run specific tasks, which you can store in the server's custom rules for future requests.
Break a complicated request into multiple parts
- e.g. ask Glue AI first to fetch data, then ask it to take action on the data in a separate request
- Avoid asking for multiple actions in a single prompt
Specify the format you want the output to be in, such as a table or list — or better yet, save your preferences in AI Rules!

Optimize your experience by configuring rules

Save time by storing default IDs in server rules

When connecting a tool via an MCP server, you often need to provide IDs that identify where or how an action should happen — for example, which project, board, database, or repository to create something in. By saving these IDs in your server's rules, you can save yourself the step of manually providing IDs each time — Glue AI will do it for you.

Here’s how to discover the necessary IDs:

Read the tool's MCP manifest or docs

Most MCP-exposed tools will document their parameters. Look for:

Required fields (e.g., project_id, repo_name, board_id)
Default-able fields (i.e., things you can pre-fill with a rule)
Example payloads

Use the tool in Glue to see what it asks for

Try manually triggering the tool from a thread. If Glue prompts you for a value, that’s a strong hint you can set it ahead of time.

Example: If Glue AI ask “Which Linear project ID should this go into?”, that means you could write a rule that injects "project_id": "proj_abc123".