AI Governance & Security Vendor Wiki

6 min read

MintMCP

Researched 2026-06-28. Primary category: mcp-gateway. Young (publicly launched Feb 2026), venture-backed; product detail is good, financials are thin.

One-liner — A managed enterprise gateway that sits between AI clients (Claude, Cursor, custom agents) and MCP servers, hosting the servers, handling SSO/credentials, enforcing per-role tool access, and logging every tool call.

What it does

MintMCP is an MCP gateway plus agent monitor. Instead of every developer installing and configuring MCP servers locally with their own credentials, MintMCP hosts the MCP servers centrally and puts a control plane in front of them. It offers an “MCP store” of pre-approved/hosted connectors (100+ integrations claimed — Snowflake, BigQuery, Databricks, Slack, Teams, Google Drive, SharePoint, Confluence, custom APIs), wires them to enterprise SSO, and grants access by role.

Two products:

  • MCP Gateway — authentication, credential management, hosting, and access control for MCP servers. Uses Virtual MCP Bundles: curated sets of servers/tools mapped to org roles (e.g. Engineering → GitHub/Linear/Datadog; Sales → Salesforce/HubSpot), with SCIM group membership auto-granting the right bundle.
  • Agent Monitor — real-time visibility into agent behavior (file operations, command execution, tool calls) with guardrails that detect PII exposure, credential leakage, risky bash commands, and prompt-injection attempts, and can block/flag/alert.

The pitch is closing the “shadow AI” gap — developers running unmonitored local agents with broad credentials — by giving each agent its own scoped, rotatable identity and routing all tool traffic through one auditable choke point.

Where it sits in the stack

Primary category: mcp-gateway (layer: model-prompt / agent tool-access control). It’s a policy and identity layer for agent→tool calls, not a model firewall.

Lethal-trifecta role — it touches all three legs, though as a control point rather than deep content inspection:

  • Egress / sensitive-data exfiltration — granular authorization (read-only DB access, blocking CRM bulk exports), plus PII/credential-leak detection on tool outputs; can hand off to Bedrock Guardrails, GCP DLP, or Microsoft Purview.
  • Untrusted input — prompt-injection detection in Agent Monitor; inline policy via JavaScript sandboxes.
  • Sensitive data access — per-agent scoped credentials and role-based tool sets limit blast radius.

Trust zones: it’s the broker between the agent (yellow) and internal systems / SaaS (green), and it brokers untrusted tool results back (red).

Deployment & architecture

  • Managed SaaS is the default: MintMCP hosts the MCP servers (including STDIO-based servers) and manages their lifecycle and credentials. Self-hosted is offered via enterprise engagement.
  • Acts as an inline proxy between AI clients and MCP servers; every request is authenticated, routed, policy-checked, and logged.
  • Identity/SSO: OAuth 2.0, SAML, SSO; integrations named for Okta and Azure AD; SCIM for group-driven RBAC.
  • Audit: structured logs (timestamp, agent/user attribution, tool call, policy decision); retention 90 days–7 years; export to SIEM (Sentinel, Splunk, S3).
  • Guardrail integrations: AWS Bedrock Guardrails, GCP DLP, Microsoft Purview.
  • Compliance: SOC 2 Type II audited; HIPAA with BAA available; encryption in transit/at rest; data-residency options; uptime SLAs (per vendor; not independently verified here).

Positioning & differentiators

MintMCP positions as managed-and-compliant first: you get governed MCP access without operating connector infrastructure yourself. That’s the main axis separating it from neighbors:

  • vs ibm-contextforge — ContextForge is self-hosted, Apache-2.0 open source you run and customize; MintMCP is the hosted, SOC-2/SSO/SCIM-out-of-the-box alternative (vendor’s own comparison; read as marketing).
  • vs docker-mcp-gateway — Docker’s gateway is a developer/runtime packaging-and-isolation tool; MintMCP is an org-level governance/identity layer.
  • vs obot — the seed bundled “Obot / MintMCP”; both are MCP-gateway plays, but they are separate companies (Obot is its own project/vendor). MintMCP leans managed-SaaS + agent monitoring; confirm Obot’s posture on its own page.
  • vs agentgateway — agentgateway is open-source data-plane proxy tech; MintMCP is a managed product with a console, store, and monitoring.
  • vs natoma, arcade, pomerium — overlapping MCP access-control / agent-identity space; differentiators are managed hosting, Virtual MCP Bundles, and the bundled Agent Monitor.

Differentiator to highlight: Virtual MCP Bundles + SCIM-driven role bundles, and the bundled agent-behavior monitor (not just a gateway).

Ownership, funding & M&A

  • Independent venture-backed startup. Legal entity reported as Dependable AI, Inc.
  • Founders: Jiquan Ngiam (co-founder & CEO; ex-Google Brain, ex-Coursera), Vijay Vasudevan (ex-Google), Bozhi “Bo” See (ex-Brilliant Home, Shopkick). Founders described as early Google Brain members; advised by Andrew Ng.
  • Backers reported: Coatue, Hustle Fund, Maven Ventures, WVV Capital; angels incl. Andrej Karpathy, Jeff Dean, Scott Belsky. Raise amount and round not disclosed in a primary source seen — left blank.
  • Public launch: 2026-02-05 (press release).
  • HQ: San Francisco (press/LinkedIn). Founding year not confirmed — left blank.
  • No M&A; no acquisition flagged in seed. (Seed only bundled the name with Obot.)

CTO / hedge-fund lens

  • Day-2, conditional on Day-1. You only need an MCP gateway once agents are actually calling internal tools/data. For a fund, MintMCP becomes relevant when desk/quant/ops teams start wiring Claude/Cursor agents to Snowflake, SharePoint, Slack, internal APIs — at which point ungoverned local MCP credentials are a real exfiltration and audit problem.
  • Fit: medium. The managed model is attractive for a small IT/security team (no connector infra to run), and SOC 2 / SSO / SCIM / SIEM-export / 7-year retention map cleanly onto financial-sector audit and surveillance expectations. Caveats: (1) it’s a new company (launched 2026) — diligence the SOC 2 report, data residency, and roadmap stability; (2) managed SaaS means MintMCP hosts your connectors and brokers credentials — that concentration may be a non-starter for the most sensitive data unless self-hosted; evaluate the self-hosted option. Guardrail/PII claims lean partly on third-party engines (Purview, GCP DLP, Bedrock) — verify what’s native vs delegated.
  • Not a model-risk / SR 11-7 tool; it’s access governance and audit plumbing.

Competitors / alternatives

obot, ibm-contextforge, docker-mcp-gateway, agentgateway, natoma, arcade, pomerium

Open questions / to verify

  • Funding round, amount, and date (not disclosed in a primary source as of 2026-06-28).
  • Founding year and exact incorporation (Dependable AI, Inc.).
  • Self-hosted deployment details — is it full feature parity, and where do credentials live?
  • Native vs delegated guardrails: which detections are MintMCP’s own vs Purview/GCP DLP/Bedrock.
  • Relationship/overlap with obot beyond the seed’s bundling — confirm they are unrelated companies.
  • SOC 2 Type II report scope and any independent corroboration of compliance claims.

Sources

History

  • [2026-06-28] Stub created from seed registry.
  • [2026-06-28] Researched; established MintMCP as an independent venture-backed enterprise MCP gateway + agent monitor (managed SaaS, self-host optional), legal name Dependable AI Inc., SF, founders Jiquan Ngiam/Vijay Vasudevan/Bozhi See, publicly launched 2026-02-05, backers Coatue/Hustle Fund/Maven/WVV (amount undisclosed). Set ownership_confidence medium, hedge_fund_fit medium, status researched. Kept category mcp-gateway. Cached 4 sources.

Graph View

Backlinks