Skip to main content
npx -y agentlinkops agent setup writes this MCP entry for each detected client and installs the skill pack beside it; use this page when your client is not detected or does not support skills. Add https://app.agentlinkops.com/mcp to your client’s remote HTTP MCP servers. Hosted access requires a pilot invitation and an authorized workspace. You can use the hosted service without a local ledger or context server.

​
Connect and confirm access

  1. Add the endpoint in your client’s server settings.
  2. Complete browser sign-in and consent. Choose the intended account, workspace and project access.
  3. Refresh the tool list, then call list_projects.
  4. Confirm the project before an import or monitoring write.
Use this first instruction:
If no project appears, check the selected workspace and ask its owner to confirm your project access. Create a project in the app only if you have permission and intend to start a new one. Reconnecting cannot reveal a project outside your grant. The client’s credential store holds the OAuth grant. Keep the token there. HTTP and CLI use their own credential setup; account access explains grant and credential limits.

​
Understand what you connected

Hosted MCP exposes the same authenticated operations as the shared HTTP and CLI command catalog. Check current availability and the connected tool list before choosing a workflow. Provider settings can still prevent a registered operation from running. Two optional connections have different purposes:

​
Bounded views

Release ba77c919 (2026-09-16) serves explicit views of the same catalog, all from one registry and all callable for every command:
  • /mcp is the bounded default: five core commands plus search_tools, describe_tools, run_read_command and run_write_command, with every command listed by toolset in the server instructions. A client connected before this release keeps its flat listing until it reconnects; its direct calls keep working.
  • /mcp/all keeps the flat catalog for a client with its own tool search (Claude Code, Hermes).
  • /mcp/{toolset} serves one group (monitoring, evidence, discovery, library, competitors, reports, workspace, notifications, webhooks); /readonly after any view, or the X-MCP-Readonly: true header, removes writes.
  • /mcp/code serves search, describe and execute, which runs your JavaScript in an isolated sandbox with every command callable as await agentlinkops.<name>({...}), for batches and joins.
List commands answer concise rows on MCP by default (format: "detailed" restores every field; HTTP and the CLI stay detailed), and nine retired command names keep resolving as aliases for twelve months. agentlinkops agent setup writes the right view for each detected client, and the agent reference is what the installed skill has the agent read before its first call.

​
Reconnect after a failure

  • No tools: check the endpoint and HTTP transport, finish browser consent, then refresh discovery.
  • 401: reconnect the account to restore valid authentication.
  • 403: check workspace membership, project access and the operation’s required scopes. Ask the workspace administrator to correct access; repeated consent cannot widen a forbidden grant.
Once the project read succeeds, check a placement. Keep the returned project identifier with the task. Use troubleshooting for interrupted jobs, expired history or unavailable providers.