MCP Gateway
for Agents
One gateway. Govern every agent→tool call. Policy, approvals, and audit for MCP tools — designed for agent fleets and deployable on‑prem.
Why Palma.ai
Protocol-true MCP gateway built for enterprise governance, agent fleets, and on-premise deployment with complete control.
Want a personalized walkthrough? Schedule a demo with our team.
Protocol‑true to MCP
No custom one‑offs; future‑proof as the spec evolves. Keep the protocol intact to avoid lock-in and spec drift.
Agent‑first
Built for agent↔tool operations (not just human chat UIs). Designed for fleets and pub/sub namespaces.
Governance‑grade
Per‑tool policy, staged rollout, human‑in‑the‑loop approvals. Complete audit trails for compliance.
On‑prem first
Run the core gateway in your Kubernetes/OpenShift; SaaS control plane optional. Full on-prem available.
Cost safety
Usage visibility and token/cost anomaly alerts (roadmap) to prevent runaway spend.
Built with design partners
Co‑developed with regulated industries for real-world enterprise requirements.
How it works
Three simple steps to govern your entire agent fleet with enterprise-grade security and control.
Register servers
Register MCP servers in the control plane (vendor‑official or your own). We pull live list_tools so policies stay current.
One endpoint
Agents connect to one endpoint (URL or pub/sub). The gateway presents as an MCP server and forwards as an MCP client.
Enforce & route
Policy, approvals, and auth enforced at the gateway — then routed to the right upstream server with full audit.
Agent Fleet
Multiple AI agents collaborating seamlessly via the Model Context Protocol
Palma Gateway
Central governance & orchestration hub
MCP Servers
Enterprise systems exposed via Model Context Protocol
Built for agent fleets
Stateless & sessionful
Stateless mode for elasticity today; sessionful MCP support coming for advanced flows.
Pub/sub addressing
Works with pub/sub addressing (e.g., NATS FQNs) and a single global namespace.
Error handling
Standardized error envelope so frameworks can capture, recover, and re‑invoke consistently.
Key capabilities
Policy & RBAC
Default read‑only, per‑tool allow/deny, staged rollout from safe → write. Assign to roles, departments, users, and agents.
Human‑in‑the‑loop approvals
Mark sensitive tools as Require Approval. Target approvers by role/OU; integrate with your agent's pending/resume flow.
Auth mediation (OAuth2/JWT)
Human OAuth2 code flow today; service‑to‑service bearer/JWT for agents and per‑tool scopes on the roadmap, aligned with MCP.
Observability & cost safety (roadmap)
Per‑agent/per‑tool usage analytics; token/cost anomaly alerts; centralized audit at the gateway.
Deploy anywhere
Core gateway on‑prem (K8s/OpenShift/Docker). SaaS control plane optional; full on‑prem available for highly regulated orgs.
API/CLI‑first
UI for fast setup; HTTP API for automation; bring your own CLI.
Always On Spec
Use vendor‑official MCP servers (email, calendar, CRM, ITSM, data) or BYO internal servers. We keep the protocol intact to avoid lock‑in and spec drift.
Why Palma vs Other MCP gateways
See how Palma.ai compares to other MCP gateway solutions in the market.
| Criterion | Palma.ai | Other Gateways |
|---|---|---|
| Core value | Governance control plane (policy/approvals/audit) | Tool building + auth + registry |
| MCP stance | Protocol‑true gateway | SDK/TDK + registry |
| Deployment | On‑prem first or SaaS control plane | Hosted/VPC; dev‑centric |
| Governance | Per‑tool rules; staged rollout; approvals; param controls (next) | Auth + secure tool calls |
| Observability | Agent/tool usage; token/cost anomaly (roadmap) | Dev‑oriented telemetry |
Ready to see the difference? Experience enterprise-grade MCP governance firsthand.
FAQ
Does Palma replace MCP servers?
No. Palma fronts them with a single governed endpoint and keeps the protocol intact.
Can we run fully on‑prem?
Yes. Core gateway on‑prem; control plane can be on‑prem or SaaS.
Do you support agents (not just chat UIs)?
Yes. Palma is agent‑first — designed for fleets and pub/sub namespaces.
How do approvals work?
Mark tools as Require Approval. The gateway elicits approval, your agent pauses, and resumes upon approval — auditable end‑to‑end.
What about OAuth2 and scopes?
Human OAuth2 ; service‑to‑service/JWT and per‑tool scopes, aligned with MCP updates.
Is there a CLI?
We expose an HTTP API so you can bring your own CLI.
Ready to govern your agents?
Get pilot access — we'll stand up a trial (SaaS or on‑prem) and integrate your first MCP servers.