PRODUCT UPDATES

Changelog

New features and releases in Mintive (formerly MintMCP), newest first. RSS feed.

Coworker agents in Teams and your own KMS keys

Talk to your coworker agents in Microsoft Teams

Organizations that work in Microsoft Teams couldn't hand work to a coworker agent without switching to Slack. Your coworker agents now work in Microsoft Teams too: message an agent in a Teams chat and it replies in the thread with a live status card and a Stop button, running as the teammate who sent it.

Each agent's bot is created inside your own Microsoft tenant, so the install stays in your environment. It is in early access while we finish self-serve setup, so email founders@mintive.com and we will connect it for your tenant.

Docs: Coworker agents overview

Hold the encryption key to your own data

Security teams want to be able to cut off a vendor's access to their data without waiting on the vendor. You can now encrypt the connector credentials your users and agents store in Mintive with your own AWS KMS key, managed entirely in your AWS account.

Point us at your key and role from Enterprise settings, run the verification test, and enable it: credentials saved from then on are wrapped under a key only you control, and you can change or revoke it yourself. It is an Enterprise feature we turn on per organization.

Docs: Customer-managed encryption

Give auditors a read-only view of every MCP in your org

The people who audit what is connected usually shouldn't be able to change it. You can now grant a read-only, role-based view that sees every MCP across your organization: its configuration, saved tools, access lists, and full change history, with call and edit rights left off.

It is built for security and compliance reviewers. Assign it from your custom roles.

Docs: Roles and permissions

Also in this release

  • Triage what Mint Guard catches

    Mint Guard scans your agents' tool traffic for prompt injection and blocks it in enforcing mode. The new Review tab collects each high-confidence detection, so your security team can mark it benign or resolved with a note and keep an auditable record in the audit log and SIEM.

  • Custom instructions for any connector

    Write server instructions for a connector or a bundle from the tool customization tab, and Mintive serves them to the agent's client, coworker agents included.

  • Agent access maps

    Each coworker agent opens to a Visualization tab: a map of who can reach it, which triggers and Slack channels fire it, and every connector it can call with live tool counts, alongside its model, workspaces, and secrets.

  • Agent Monitor shows more

    See the actual prompt behind Claude Code and Codex runs, and who triggered each coworker agent run.

  • More in your SIEM

    LLM Gateway calls export as OpenTelemetry GenAI traces, audit logs now stream to Azure Monitor and Microsoft Sentinel, and the caller's IP lands on every activity event.

  • More coding agents governed

    OpenCode and Claude Desktop join Cursor, Claude Code, and Codex as agents Mintive governs.

  • Bedrock, GCP, and Azure, standard

    Amazon Bedrock is now a standard provider in the LLM Gateway, and you can pull connector and agent credentials from GCP Secret Manager or Azure Key Vault without asking us to switch them on.

  • Admin API with scoped keys

    Manage Mintive programmatically with bearer API keys scoped to specific permissions, against a published OpenAPI spec.

  • Require user approval, as a rule

    "Require user approval" is now a first-class rule action, so you can gate specific tool calls on a human sign-off.

  • The CLI is now mintive, and the portal can be yours

    The installed mintmcp CLI is renamed mintive (with a mintmcp alias so existing setups keep working), and organization white-label settings put your own logos across the app and the login portal.

Mint Guard counts what matters

A summary that counts every low-confidence hit makes it hard to tell how much real risk Mint Guard is catching. The Mint Guard summary now counts prompt injection only at high confidence, the level Mint Guard blocks at, and credentials and PII at medium confidence and up. Each tile says which level it counts.

Your numbers may be lower than before, but nothing changed about what Mint Guard scans or blocks. Lower-confidence findings still appear on each call's details in the Activity log.

Docs: Mint Guard

MintMCP is now Mintive

MintMCP is now Mintive. The name is new; nothing you run is. Every gateway, connector, policy, agent, and login carries over as is, on the same credentials you use today.

The Mintive wordmark and pixel logo on a green brand card.

The rename reflects what the platform has become since the MCP gateway launched: four more products now run alongside it, so this is agent infrastructure for the enterprise rather than an MCP gateway alone. The full announcement is on the Mintive blog.

Docs: Introduction to Mintive

Your own model keys and Mint Guard per connector

Run every model call through your own keys

When every agent and laptop calls models with its own key, nobody can see the spend or control which models are used. The LLM Gateway routes your agents' model calls through your own provider credentials, so prompts and responses stay on infrastructure you control and every call lands in one audit trail.

Set a monthly budget per policy or per user, then track spend against it in a usage dashboard that breaks down by person and model. Reuse the Claude subscription seats you already pay for instead of buying separate API keys, and constrain which models a policy can use. It is in early access; email founders@mintive.com and we will turn it on for your org.

Docs: LLM Gateway, Budgets

Set guardrails per connector, not just per org

One org-wide guardrail setting forces a trade-off: strict enough for risky connectors is too strict for safe ones. Mint Guard enforcement used to be a single setting for the whole organization. Now you can override it per connector from the Mint Guard page, setting request and response handling to Inherit, Off, Monitoring, or Enforcing for each one.

Every connector page shows its effective Mint Guard state, so you can tell at a glance whether a connector is being watched or blocked.

Docs: Mint Guard, Choosing guardrails

Also in this release

Agent cost tracking, GCP and Azure secrets

See what your agents cost, by person and by model

Agent bills arrive as one number, with no way to tell which people, models or sessions drove them. The Usage page in Agent Monitor now breaks agent spend down by day, model, session, and user, with estimated cost, cache hit rate, and a human versus agent split, and it opens on the last 7 days.

The Usage tab ranking users by estimated agent cost, with a top sessions list showing each session's model, tokens and cost.

Coworker agents report this by default; Claude Code and Codex laptops report once you push the telemetry settings from the Agent Monitor setup page.

Docs: Agent Monitor usage

Keep connector secrets in GCP or Azure

Copying secrets into another tool creates a second place to rotate and audit them. Mintive now resolves secrets from GCP Secret Manager and Azure Key Vault alongside AWS Secrets Manager.

Add a provider from the Secret providers page, bind it on a connector's bearer token, custom header, or Google service account form, and use Fetch keys to confirm the value resolves. GCP and Azure are available by request.

Docs: Secret providers

GitHub Copilot monitoring is on for every org

Teams that use GitHub Copilot alongside other coding agents had a gap in their monitoring. The GitHub Copilot tab is now on the Agent Monitor setup page for every organization, and Copilot appears in the agent filter on the activity view.

File access, commands, and tool calls flow into the same dashboard, rules, and SIEM export as Claude Code, Cursor, and Codex.

Docs: Agent Monitor overview, Copilot setup

Also in this release

Run your agents on Mintive

Connect your agents, tools and models through one gateway, and see every call in the log.

No sales call needed!