Skip to main content
Outcome: two tools answer from the same saved knowledge, with separate endpoint identities where appropriate. You need: a page you can read and permission to create the required connections. Have an editor available if you want to test writing.

​
1. Prepare a reference page

Create or choose an approved checklist in Company. Give it a clear title and link it from the index. Confirm it appears in Search.

​
2. Connect a file-capable tool

Pair its machine through Connections → Connect an agent. Let the setup prompt install the replica and harness skills, then check nlkb doctor. Ask the tool:
A second tool on the same machine can use Add this harness and the existing replica.

​
3. Connect a remote client

Create a separate Remote MCP connection, select the required vault, and configure the endpoint in the client’s supported connector settings. Complete OAuth or secure Bearer-key setup. Ask it to run orient, search for the same checklist, open it, and summarize with the source path. Remote MCP does not require the desktop viewer or a local folder.

​
4. Compare evidence

Both tools should identify the same page and use its current content. If answers differ, compare the source they opened, the active workspace, and the local replica’s health. Do not assume model-generated prose is itself stored knowledge.

​
5. Add writing deliberately

For a controlled test, use an editable test page and ask one tool for a small change. Inspect the saved page and Activity afterward. Do not ask both tools to edit the same stale version simultaneously; conflicts retain competing changes for review. For consistent company output, configure Company Voice and ensure each tool can read its guidance pages.

​
If it did not work

If answers differ, follow Missing agent sources and compare the actual pages opened. If an agreed write becomes a conflict, inspect its retained candidate rather than repeatedly retrying. Next: Company writing.