Release v1.0.3

tigergraph-mcp 1.0.3 Release Notes

This release gives shared deployments control over what an agent can reach, and a record of what it did.

New

• Tool-call logging, with optional caller identity — a deployment can now keep a record of what has been run against a TigerGraph instance, and optionally attribute each call to a connection profile or a TigerGraph account. Previously a shared server kept no such record. Both are off by default, and identity is a second, separate opt-in: an account name identifies a person in most deployments, so the server does not write one into an operator's logs unasked.
• Serving a subset of the tools — a deployment can restrict which tools it offers, by category, by whether they only read, or individually. This cuts what an agent carries on every request — offering only the tools that read more than halves it — and limits what an agent is able to do at all.
• Tools declare their own behaviour — every tool now tells the client whether it only reads, whether it can destroy data, and whether it is safe to repeat. Clients can run reads without interrupting the user and ask for confirmation before anything that removes data.

Changed

• A single HTTP session can be given a narrower set of tools than the server as a whole, so one shared deployment can serve clients with different capabilities. A session can only ever be narrowed, never widened beyond what the deployment allows.
• A tool selection the server cannot honour is reported at startup rather than resulting in a server that quietly offers the wrong tools.
• Documentation for HTTP mode now covers running and connecting to a server end to end, with client examples for both supported MCP SDK generations.

Upgrading

No action required. Both new capabilities are off by default: the full tool set is served unless a selection is configured, and nothing is logged about tool calls unless enabled.

Deployments that enable caller identity at the account level should note that log lines will then contain TigerGraph account names, which is personal data in most jurisdictions. Whatever retention and access rules apply to other logs apply to these.