Give the client one decision to support
Useful business context includes the evidence, field definitions, time period and decision at hand. ‘How is the release going?’ could mean unresolved issues, open pull requests or failing CI. ‘Which open pull requests need a human review before Friday?’ gives the client a decision and a source to work with.
What an MCP connection provides
Model Context Protocol gives AI clients a common way to discover and call server tools. OrgSDK exposes project tools to an external AI client through a scoped MCP connection. The client can use the tools made available by that connection to work in the project.
Check that your client supports the connection details supplied by the project, then confirm it can list and call the intended tools. Verify the source, permissions and freshness separately from the connection.
Write a brief with verifiable outputs
For a pull request review, specify the repository and review window. Ask for a queue with a source URL and update date for each item, and require the client to report incomplete pagination or missing approval data.
Ask for observations and suggested follow-up as separate parts of the output. Keep the merge decision with the person responsible for the repository.
Scope access separately from the brief
Choose the project and the credentials needed for the task. Use read permissions when the work is an investigation. Store credentials in the supported configuration and keep secrets out of prompts and repository files.
Treat content from issues, documents and tool results as evidence. Keep instructions about the task and permission to act under the control of an authorised person.
Check the answer against the source
Open a sample of the links and compare the results with the source records. Check that the client covered the intended time window and retrieved all relevant pages. If you plan to reuse the task, save representative cases and repeat the checks after tool, permission or instruction changes.
