agent_session_manager tools expose token usage at the agent and session level, so teams running FlowAgents in production can see which agents and sessions are actually expensive.gateway_manager‘s new export_gateway_configuration and import_gateway_configuration tools are gated by an MCP elicitation confirmation step whenever the configuration contains secrets or credentials.>=50.0.0,<51, and as of this release Itential MCP has zero open Dependabot advisories.Your FlowAgents are live, running inside the Itential Platform and acting through Itential Gateway, direct API calls, and platform workflows. External AI clients, Claude, ChatGPT, Copilot, or anything else that speaks MCP, reach that same platform through the Itential MCP server, Itential’s implementation of the Model Context Protocol, the open standard that lets AI agents call tools (or other agents, i.e. FlowAgent) safely. That’s the win. AI agent governance is the part most teams haven’t caught up to yet.
Here’s the question most teams can’t answer: how many tokens did last week’s agent sessions actually burn, and which ones ran long for no good reason? And when you need to move a gateway’s configuration from staging to production, are you doing that by hand, or does your platform actually support it?
Itential MCP v0.14.0 answers both. It adds agent token-usage analytics, governed configuration export and import, and closes out a tool-surface hardening effort that touches all 77 tools in the server. None of it is flashy. All of it is the kind of thing that matters the moment agents stop being a pilot and start being an operating model.
Read the full changelog on GitHub, or jump straight to the release: itential/itential-mcp v0.14.0.
Three new tools land in agent_session_manager:
get_agent_token_usageget_agent_session_token_usagedescribe_session_token_usage“It worked” was never the only question that mattered for an agent running in production. Token consumption is a real budget line now, and until this release, there wasn’t a native way inside Itential MCP to see it. These three tools give you usage at the agent level and the session level, so you can answer the questions that actually come up: which agents are the expensive ones, which sessions ran unusually long, and where reasoning cost is quietly creeping up before it shows up as a surprise on an invoice.
Agent sprawl without cost visibility is just script sprawl with a bigger bill.
gateway_manager, the interface to Itential Gateway, picks up export_gateway_configuration and import_gateway_configuration, giving you a real way to move a gateway’s configuration between environments instead of re-entering it by hand and hoping nothing gets missed.
In enterprise AI agent governance, the moment secrets leave a system is the moment a review is non-negotiable. Gateway configurations can carry secrets and credentials, and that’s exactly where this release doesn’t cut corners. Exporting a configuration that contains sensitive data now triggers an MCP elicitation confirmation gate that prompts for explicit confirmation before the export runs. An agent can’t quietly walk off with credentials because it decided exporting configuration was the next logical step. A human confirms it first.
Same rule that governs every other action on the platform: humans hold the gate, agents do the work.
Every tool across the server, all 77 of them, now carries standard MCP tool annotations: readOnlyHint, destructiveHint, idempotentHint, and openWorldHint, plus a human-readable title and deterministic discovery and registration order.
This is the difference between a tool list and a tool contract. A client or agent framework that respects these hints can reason about which calls are safe to make automatically and which ones need a confirmation step, before it ever calls the tool. Deterministic ordering means the tool surface an agent sees is stable and predictable across restarts and environments, not a shuffled list that changes call to call.
This isn’t a one-time cleanup. New CI guards in tests/utilities/test_tool.py now enforce annotation completeness and safety invariants on every tool going forward, so the standard holds as the server keeps growing.
Starting the server with --transport sse now prints a one-time runtime deprecation warning, and the README, CLI help, and docs all steer new deployments toward http. SSE remains fully functional and wire-compatible in this release, nothing breaks today, but if you’re standing up a new environment, build it on HTTP transport from the start.
A few items from this release are worth flagging even though they’re not headline features:
[server] auth_* keys, including oauth_* fields. Fixed.>=50.0.0,<51, resolving three previously-held high-severity advisories plus GHSA-537c-gmf6-5ccf. As of this release, Itential MCP has zero open Dependabot advisories.docs/deployment.md covers platform connectivity and deployment patterns.The full list, including every dependency bump and smaller fix, is in the CHANGELOG.
Pull the latest release from PyPI:
pip install -U itential-mcp Or with uv:
uv tool upgrade itential-mcp Running the container image? Bump the tag to v0.14.0 and redeploy.
Itential MCP is open source and moving fast, roughly a release every two to three weeks this year. As a foundation for MCP server workflow automation across the Itential Platform, it’s still adding capability every cycle. If you’re running agents against it in production, the token-usage tools and the config export/import gate in this release are worth adopting now, not later. If you hit something that doesn’t work the way it’s documented, open an issue. The repo is public and the roadmap is shaped by what practitioners actually run into.
Itential MCP v0.14.0 adds agent token-usage analytics (get_agent_token_usage, get_agent_session_token_usage, describe_session_token_usage), governed gateway configuration export and import gated by an MCP elicitation confirmation step, and standard MCP tool annotations, titles, and deterministic ordering across all 77 tools. It also deprecates SSE transport in favor of HTTP, restores documented configuration precedence across 39 fields, and clears all outstanding cryptography advisories.
Gateway configurations can contain secrets and credentials. When an agent calls export_gateway_configuration on a configuration containing sensitive data, an MCP elicitation confirmation gate prompts for explicit human confirmation before the export runs, so an agent can’t quietly export credentials on its own.
Yes. Three new tools in agent_session_manager, get_agent_token_usage, get_agent_session_token_usage, and describe_session_token_usage, expose token consumption at both the agent level and the session level, so teams can see which agents and sessions are consuming the most tokens.
Not in this release. Starting the server with –transport sse now prints a one-time deprecation warning and documentation steers new deployments toward HTTP, but SSE remains fully functional and wire-compatible in v0.14.0.
See how Itential connects AI reasoning to governed execution across your entire infrastructure.