Agent Custom Instructions
OneMCP exposes many MCP tools through a compact discovery layer (find_relevant_tools, call_tool, list_mcp_connections). Add a short custom instruction so your AI client reliably uses that flow instead of stopping at "I don't have a tool for that."
These instructions steer client behavior. They do not connect OneMCP - an admin still adds the OneMCP endpoint to the client. See OneMCP for the discovery flow and the Integration Guide for connection setup.
The instructions
Paste this block into whichever custom-instruction surface your client supports (see placement below). It is under 10 lines and works the same across clients.
You have access to Quilr OneMCP, a gateway that exposes MCP tools through a discovery layer.
For anything that needs external data or actions, use these tools instead of guessing:
1. Call find_relevant_tools with a short description of the task before concluding a capability is unavailable.
2. Call call_tool with the exact tool name and arguments returned by the search.
3. If a tool reports it needs authentication or a connection, call list_mcp_connections so the user can connect or reconnect, then retry the request.
4. Use save_or_update_memories and search_memories to store and recall durable user context.
Only act on results returned by call_tool. Never invent tool output.
Ready-to-use files
If you prefer not to copy-paste, download the packaged versions:
- Claude Skill (
SKILL.md) or the ready-to-upload.zipfor organization provisioning. copilot-instructions.mdto drop into a repo's.github/folder.
Where to add it
Claude
- Per user: Settings > Profile custom instructions, or a Project's instructions.
- Org-wide (Team/Enterprise): Organization settings > Organization and access > Organization instructions (max 3,000 characters), or provision the Claude Skill as an organization Skill. Applies to all members and takes precedence over individual user instructions.
ChatGPT
- Per user: Settings > Personalization > Custom instructions, or a Project's instructions.
- Org-wide: ChatGPT has no setting that pushes custom instructions to every user's chats. Publish a shared Project or a workspace GPT that carries the text; it applies only inside that Project or GPT.
GitHub Copilot
- Per repo (covers the IDE): add
.github/copilot-instructions.mdwith the text. Read by Copilot in VS Code, JetBrains, the CLI, and on github.com. - Org-wide (Business/Enterprise): Organization Settings > Copilot > Custom instructions. This covers Copilot Chat on github.com, code review, and the coding agent only. It does not reach the VS Code / JetBrains IDE or the CLI, so ship
.github/copilot-instructions.mdin each repo for IDE and CLI coverage.