Managed MCP Servers
Hosted Model Context Protocol servers — ready-to-connect tool providers for AI agents
The Model Context Protocol (MCP) defines how AI agents discover and invoke external tools. Running those tool servers yourself means managing more infrastructure. We host a growing catalog of MCP servers and operate custom ones — your agents connect, we keep the servers running.
What MCP Servers Enable#
MCP servers expose tools that agents can call during a task: query a database, search the web, read a file, send a notification, or call any API. Instead of hardcoding integrations into each agent, tools are declared by the server and discovered at runtime.
Benefits for agent developers:
- Tool definitions are versioned and discoverable — no manual schema updates
- Authentication handled at the server level, not in agent code
- New tools available to all agents without redeployment
- Rate limiting and quotas enforced at the server, not per-agent
Pre-Built Server Catalog#
Data & Databases#
| Server | Capabilities |
|---|---|
| PostgreSQL | Query, insert, update, describe schema, run migrations |
| MySQL / MariaDB | Read/write queries, schema inspection |
| MongoDB | Document CRUD, aggregation pipelines |
| Redis | Key-value get/set, pub/sub, stream operations |
| ClickHouse | Analytical queries, time-series data |
| Elasticsearch | Full-text search, document indexing |
Search & Web#
| Server | Capabilities |
|---|---|
| Web Search | Google, Bing, and Brave search with configurable result count |
| Web Scraper | Fetch and parse any URL, extract structured content |
| News | Current events from curated sources with date filtering |
| Academic Search | Papers from ArXiv, Semantic Scholar, and PubMed |
Code & Compute#
| Server | Capabilities |
|---|---|
| Python Sandbox | Execute Python code in an isolated environment |
| Node.js Sandbox | Execute JavaScript/TypeScript in an isolated environment |
| Shell | Run bash commands in scoped containers |
| Git | Clone, diff, log, and read repository content |
Files & Storage#
| Server | Capabilities |
|---|---|
| File System | Read, write, list, and search files in scoped directories |
| S3 / GCS | Object storage read/write, presigned URLs |
| Google Drive | Document read, search, and creation |
| Notion | Page read, write, and database queries |
Communication & Productivity#
| Server | Capabilities |
|---|---|
| Slack | Send messages, read channels, manage threads |
| Send via SMTP/API, read inbox (OAuth) | |
| Calendar | Create events, check availability, invite attendees |
| Linear | Issue CRUD, project queries, status updates |
| GitHub | Issues, PRs, repository operations, workflow triggers |
Custom MCP Server Hosting#
Have your own MCP server? We host it with the same reliability SLA as our catalog servers.
What we provide:
- Container hosting with health checks and auto-restart
- TLS certificate management and HTTPS endpoint
- Authentication layer (API keys or OAuth — your choice)
- Horizontal scaling based on request volume
- Request logging and error alerting
- Rolling deploys with zero downtime
Supported runtimes: Node.js, Python, Go, and any containerized binary.
Deployment: Push a container image or connect a Git repository and we build from source on every release tag.
Security Model#
Authentication#
Every MCP server connection requires an API key scoped to a specific agent or team. Keys rotate automatically on a configurable schedule.
Tool-Level Permissions#
Restrict which tools an agent can invoke even within a connected server:
1{2 "agentId": "research-agent",3 "server": "postgresql",4 "allowedTools": ["query", "describe_schema"],5 "deniedTools": ["insert", "update", "delete", "drop"]6}Network Controls#
- IP allowlist per server
- Private network access (VPC peering available)
- Egress filtering on sandbox servers
- mTLS for high-security environments
Audit Log#
Every tool call recorded with: agent identity, tool name, input arguments (redacted fields configurable), output size, latency, and timestamp. Exportable to your SIEM.
Reliability#
| Metric | Commitment |
|---|---|
| Availability | 99.9% uptime SLA |
| Response latency | p99 < 500ms for database and search servers |
| Sandbox cold start | < 2 seconds |
| Failover | Automatic region failover for critical servers |
Pricing Model#
- Catalog servers — charged per tool call, no base fee
- Custom hosting — flat monthly rate per server based on resource profile
- Enterprise — reserved capacity with predictable pricing
All plans include request logging, alerting, and the management dashboard.
Getting Started#
Connect your agents to the first MCP server in under 10 minutes. No infrastructure setup, no TLS configuration — just an endpoint and an API key.
Get your MCP API key →