Mintive

• An identity, network, or data platform already licensed by the buyer could add MCP tool gating and per-agent tokens.
• Platform fees that scale with usage and team size could raise a customer's bill as it adds agents, which could slow expansion inside an account.
• Mintive calls Coworker Agents, a Slack agent product still in beta, its fastest-growing product, so its growth could shift toward work that falls outside the security team's budget.

Mintive sells to companies whose staff already connect AI agents to internal systems through MCP, an open-source standard for linking AI applications to external systems. The buyer is the security, IT, or platform team that has to know what each agent can reach and what it did. Mintive frames the starting point as trust, saying a company needs to know what agents can reach, what they did, and why they were allowed to do it.

The customer accounts that describe the pain are published on Mintive's site. Coursera's story says no one could say exactly what was connected and that API tokens were scattered across the company. Stability AI's story describes workflows held together by personal API tokens in a dotenv file on someone's laptop.

Those accounts make the buyer and the problem concrete. No independent survey or incident count in the reviewed sources measures how widespread the problem is, so every account of its scale in the record is Mintive's or a customer's. [s4, s23, s13, s26]

The MintMCP Gateway, which Mintive now markets as its MCP Gateway, is the core product. It hosts and catalogs approved MCP servers and groups them into role-based bundles behind single sign-on. An administrator can switch off individual tools, such as destructive ones. It logs each call with the user, tool, server, and guardrail result for export to a SIEM.

Three products extend that gateway. Agent Gateway gives each agent its own token, with access delegated from a user, a team credential, or a service account, while the gateway keeps the raw credentials. Agent Monitor logs MCP calls, bash commands, and file access and replays sessions call by call. Mint Guard, a detection policy Mintive maintains, screens tool calls and results for prompt injection and, when set to enforce, blocks high-confidence matches.

Two newer products are in early release. The LLM Gateway, which routes agents' model calls and sets per-team budgets, is in early access. Coworker Agents runs agents in Slack with memory kept in a GitHub repository, and it is in beta on the Enterprise and Teams plans. [s5, s6, s7, s31, s9, s10, s2]

Mintive competes with other MCP gateways for the same security and platform buyer. Deerfield Group says, in a Mintive customer story, that a partner connected it with Runlayer, which it evaluated before choosing Mint. Natoma, another MCP gateway vendor, says on its platform page that it is joining forces with Snowflake.

Identity and network vendors sell into the same layer. SiliconANGLE's report on Okta's Cross App Access ecosystem lists MintMCP in one group with Cloudflare, Stytch, and WorkOS. SiliconANGLE describes that group as routing and securing the traffic between agents and the apps they reach. A buyer that already licenses one of those vendors can compare Mintive's controls with what that vendor offers.

Mintive adds per-agent identities, inline screening, and model-call budgets to the gateway. Its middleware can send tool calls through AWS Bedrock guardrails, Google Cloud Model Armor, or OpenAI moderation, so a customer keeps the data-loss tools it already runs. [s25, s24, s18, s6, s8, s9]

Mintive names its customers and publishes accounts from several of them. On its homepage, Mintive lists names above a link to its customer stories, including Coursera, Braze, Stability AI, Eurostar, and Deerfield Group. Its About page says teams at Coursera, Braze, DNSFilter, Workstream, and Flashfood use Mint to govern their agents.

The customer stories report deployment figures. Coursera's story cites about 2,000 employees on Mint and more than 70 MCP connections, Stability AI's more than 140 employees, and Deerfield Group's 80 users with plans to pass 200 people. Mintive publishes these accounts on its own site.

Two outside sources name Mintive's partnerships. Cursor lists MintMCP among its hooks partners in a December 2025 post, and SiliconANGLE's June 2026 report places it among the gateway providers in Okta's Cross App Access ecosystem.

Mintive sells through demos and custom quotes. Its pricing page names per-user licensing based on active AI agent users plus platform fees that scale with usage and team size. Its quote form offers team sizes from 1 to 100 through more than 10,000. [s3, s4, s12, s13, s26, s25, s21, s18, s11]

Mintive's founders come from AI research and large-scale engineering. Mintive says Jiquan Ngiam and Vijay Vasudevan were early members of Google Brain who helped build TensorFlow and Coursera. Its About page also lists Bozhi See and Phillip Bensaid, with prior roles at Brilliant Home, Shopkick, Disney+, and Amazon.

The company started as Lutra AI. TechCrunch's December 2023 article describes Ngiam's years at Coursera and Google and the seed round Lutra closed that year, when it was in private beta with a small number of customers.

That record shows named engineering builds in AI and education. The reviewed sources document no earlier security or identity product from the founders. [s4, s17, s22]

Mintive presents standard enterprise trust documentation. It says an independent third party audits it annually for SOC 2 Type II with monitoring through Drata, that it completed a HIPAA attestation, and that it signs business associate agreements. Its SafeBase trust center lists CASA Tier 2, HIPAA, SOC 2 Type 2, and an application penetration test.

The documentation matters because the gateway stores credentials and the record of each agent action. A security review can check these reports before the buyer routes agent traffic through Mintive.

The attestations ease procurement without blocking a substitute. A funded rival can obtain the same SOC 2 and HIPAA documentation, and the cited record identifies no mandate that requires a governed MCP gateway. [s8, s4, s14]

Add analyzed competitors to compare them side by side with Mintive.

Mintive sells to the security, IT, or platform team that governs AI agents the rest of the company already uses. Mintive says its agent identities work with Claude, Cursor background agents, and ChatGPT workspace agents. Its About page says a company needs to know what agents can reach, what they did, and why they were allowed to do it. Mintive renamed itself from MintMCP in October 2026 and says everything customers run today carries over to the new name.

The named customers span company sizes. Mintive labels its Coursera story Enterprise and its Workstream story Midmarket, and its quote form offers team sizes from 1 to 100 through more than 10,000.

Mintive frames agent adoption in four levels. Its About page says the gateway moves a company from level 1 to level 2 with controls in place, and that monitoring and Slack-based coworker agents are live as the next steps.

Mintive's capabilities center on holding the connection between an agent and the tools it calls. The MintMCP Gateway hosts and catalogs approved MCP servers and groups them into role-based bundles behind single sign-on. It switches individual tools on or off and logs each call with the user, tool, server, and guardrail result. Agent Gateway gives each agent its own short-lived tokens over credentials the gateway keeps, delegated from a user, a team, or a service account.

Mint Guard adds inline screening. Mintive maintains the detection policy, which screens both the arguments an agent sends and the results a connector returns. When enforcing, Mint Guard blocks prompt injection at high confidence and logs other detections, such as credentials and secrets. It can also be turned off. Rules and middleware run on the same tool traffic, and middleware can route calls through AWS Bedrock guardrails, Google Cloud Model Armor, or OpenAI moderation.

No data or model asset behind these features appears in the reviewed sources. Mintive's documentation does not say what the Mint Guard policy is built from. Customers configure role bundles, and their audit logs contain agent activity records. MCP is an open-source standard. Mintive's advantage in the record is the gateway's position in each customer's agent traffic.

Mintive names its customers and publishes stories from several of them. Its About page says teams at Coursera, Braze, DNSFilter, Workstream, and Flashfood use Mint to govern their agents, and its customer stories describe deployments at Coursera, Deerfield Group, Stability AI, and Workstream. Coursera's story cites about 2,000 employees on Mint and more than 70 MCP connections, and Stability AI's cites more than 140 employees.

Partners outside Mintive name the company too. Cursor lists MintMCP among its hooks partners in a December 2025 post and describes how it inventories MCP servers and scans responses through Cursor hooks. SiliconANGLE's June 2026 report lists MintMCP among the gateway providers in Okta's Cross App Access ecosystem.

The disclosed capital is a 3.8 million dollar seed round from 2023, when the company operated as Lutra AI. TechCrunch reported that Coatue, Hustle Fund, Maven Ventures, and WVV Capital joined the round with angels including Andrej Karpathy, Jeff Dean, and Scott Belsky.

Mintive prices through a custom quote. Its pricing page names per-user licensing based on active AI agent users, platform fees that scale with usage and team size, self-hosted deployment as an option, and enterprise SLAs and dedicated support.

The meter ties revenue to agent adoption inside an account. Each additional active agent user adds a license, and platform fees rise with usage, so a customer that rolls the gateway out company-wide pays more as use grows.

Packaging ties some products to specific plans. Mint Guard is available on the Enterprise plan, and Coworker Agents is in beta on the Enterprise and Teams plans.

Mintive delivers software that the customer's team configures and operates. The customer chooses the approved servers, the role bundles, the agent credentials, and the Mint Guard mode, and reads the audit log. Mintive maintains the Mint Guard detection policy, and its pricing page lists self-hosted deployment among its options.

Agent traffic routed through the gateway depends on it. Tool calls authenticate through the customer's identity provider, and with Mint Guard set to enforce, a call it blocks is stopped and returns an error to the agent. Every call is logged at the gateway. For local coding tools Coursera uses hooks that report activity instead of routing all traffic through the gateway.

Mintive's newer products extend what it operates for customers. Each Coworker Agents run starts on Mintive's hosted runner in an isolated sandbox with restricted network egress. The agents' instructions and memory stay as files in the customer's GitHub repository.

Mintive presents the trust documentation an enterprise security review asks for. It says an independent third party audits it annually for SOC 2 Type II, with continuous monitoring through Drata, that it completed a HIPAA attestation, and that it signs business associate agreements. Its SafeBase trust center lists CASA Tier 2, HIPAA, SOC 2 Type 2, and an application penetration test, and Mintive offers EU and US data residency.

That documentation matters because the gateway stores credentials and the record of each agent action. Mintive says its audit records are immutable and exportable to a SIEM, which gives a buyer's security team its own copy of agent activity.

The attestations ease procurement without blocking a replacement. A funded rival can obtain the same SOC 2 and HIPAA documentation, and the cited record identifies no mandate that requires a governed MCP gateway.

Mintive positions its gateway as one control point for every AI client a company uses. Mintive keeps governance in the gateway rather than in any one AI client, so a team can change AI tools without redoing its access policy, credentials, or audit trail. The same users and groups carry over to its LLM Gateway, which sets per-team budgets and allowed models.

Partnerships tie Mintive to AI client and identity vendors. Cursor names MintMCP as a hooks partner, Mintive labels Okta and Cursor as official partners on its homepage, and SiliconANGLE places MintMCP in Okta's Cross App Access ecosystem.

SiliconANGLE's report also puts Mintive in a group with identity and network vendors. SiliconANGLE's group of gateway and identity providers includes Cloudflare, Stytch, and WorkOS, and Natoma, a rival MCP gateway, says it is joining forces with Snowflake. A data or identity platform could add tool gating to a suite its customers already license.

Mintive's founders bring AI research and large-scale engineering. Mintive says Jiquan Ngiam and Vijay Vasudevan were early members of Google Brain who helped build TensorFlow and Coursera. Its About page lists Bozhi See and Phillip Bensaid with prior roles at Brilliant Home, Shopkick, Disney+, and Amazon.

The company began as Lutra AI, which TechCrunch covered in December 2023 when it closed its seed round and was in private beta with a small number of customers. Ngiam announced the Mintive name on October 12, 2026.

The documented builds are in AI and online education. The reviewed sources show no earlier security or identity product from the founders.