Introduction to Mintive
Run AI agents in production with control over what they can reach, the models they call, and what they cost. Mintive (formerly MintMCP) governs both halves of agent traffic, tool calls and model calls, so you can stop unsafe actions before they run, keep a record of what every agent did, and deploy agents that do real work alongside your team.
Add Mintive's documentation to Claude, ChatGPT, or your AI client so it can look things up while you work.
The problem
Putting agents to work takes more than connecting tools. Without visibility and control over what agents reach and spend, security teams can't approve deployment and compliance can't produce audit trails.
| Issue | Impact |
|---|---|
| No telemetry | You don't know which tools agents call or which models they use |
| Scattered credentials | API keys end up on developer laptops |
| No access control | Agents reach any connected system without oversight |
| Unchecked tool calls | Prompt injection and sensitive data pass through tool calls unscreened |
| Untracked model spend | No per-team budgets or limits on which models agents can call |
| Missing audit trails | Can't satisfy SOC 2, HIPAA, or internal risk reporting |
How Mintive helps
Mintive brings five products under one platform, so the groups, identities, and audit trail you set up once apply to every agent:
| Product | What it does |
|---|---|
| MCP gateway | Approve the MCP servers your team can use and bundle tools by role, so agents reach only what you permit |
| LLM gateway | Route model calls through Mintive with per-team budgets, allowed models, and multiple providers |
| Mint Guard | Screen every tool call for prompt injection and sensitive data, using a detection policy Mintive maintains |
| Agent Monitor | Trace every call to a person, agent, server, and model, and export the record to your SIEM |
| Coworker Agents | Run AI teammates in Slack with long-term memory, scoped permissions, and their own identity |
MCP gateway
Deploy MCP servers that your whole organization can use immediately. Instead of each developer configuring servers locally, admins curate a catalog of approved servers with pre-configured credentials and turn off destructive tools.
| Feature | Benefit |
|---|---|
| MCP store | Users browse approved servers and connect with one click |
| Pre-configured credentials | Servers work immediately after SSO, with no setup |
| Role-based tool sets | Sales teams see CRM tools, engineers see code tools |
| Hosted connectors | Run MCP servers in Mintive's infrastructure, with no local installs |
| Agent identities | Autonomous agents get their own credentials and audit trail instead of borrowing human logins |
| Centralized governance | Audit logs, access policies, and credential management in one place |
→ Gateway architecture · Quickstart · Agent identities
LLM gateway
Route Claude Code, Codex, Claude Desktop, and Coworker Agents through the LLM gateway, and every model call appears with its usage, cost, model, and latency next to the agent's tool calls. Give each team its own budget and list of allowed models using the groups you already manage in Mintive, and connect OpenAI, Anthropic, or Fireworks without rewiring each agent.
→ LLM gateway overview · Budgets · Providers
Mint Guard
Mint Guard screens every MCP gateway tool call in real time for prompt injection and sensitive data. Mintive maintains the detection policy, so your team doesn't write rules. Start in monitoring mode to see what would be blocked, then switch to enforcing. For custom scanning logic before and after tool calls, add middleware.
→ Mint Guard · Compare guardrails
Agent Monitor
See what agents like Claude Code, Cursor, GitHub Copilot, and Codex are doing. Hook scripts capture file reads, command execution, and MCP tool calls before they happen, so you can track spend by user, model, and tool, review any session call by call, and create rules that block risky behavior. Export the record to your SIEM or stream it over OTLP.
→ Agent Monitor overview · Security · SIEM export
Coworker Agents
Coworker Agents work like teammates: they live in the Slack channels your team already uses, run on schedules, and carry memory from one run to the next in files your team can read and edit. What a coworker can access depends on the channel it's in and who's messaging it, and each coworker has its own identity, so its tool and model calls run through the same gateways and Agent Monitor as the rest of your agents.
→ Coworker Agents overview · Setup · Security
What is MCP?
Model Context Protocol (MCP) is an open standard for connecting AI assistants to external tools and data sources. When an AI assistant needs to query a database, read files, or call an API, it does so through MCP.
Mintive supports MCP servers for:
- Data warehouses: Snowflake, BigQuery, Databricks
- Communication: Slack, Teams, email
- Documents: Google Drive, SharePoint, Confluence
- Custom tools: Your own internal APIs
Deployment
Mintive is available as a managed cloud service. For on-premise or self-hosted requirements, contact us to discuss your needs.
Next steps
- Quickstart: Get your first MCP server running
- Get access: Request enterprise deployment