Connections#
Lens Agents connects agents to enterprise systems through three mechanisms. All connections are governed by policies, audited, and use server-side credential injection.
Three connection mechanisms#
Native integrations#
Deep, pre-built access to core infrastructure:
- Kubernetes — full kubectl access with agent identity via K8s API impersonation. Tunnel mode or direct relay. Short-lived JWT tokens.
- AWS — EKS cluster discovery, STS AssumeRole for temporary credentials, CloudTrail attribution via session tags.
- GitHub — repository access, PRs, issues, file management via GitHub CLI.
Native integrations provide the deepest access and the tightest credential isolation.
Policy-controlled proxy#
Access any system with an HTTP/HTTPS endpoint — CRM, ticketing systems, databases, internal APIs, SaaS services. The sandbox proxy governs outbound traffic per policy, injecting credentials server-side.
If it has an API, agents can reach it — governed.
MCP server registry#
Register upstream MCP servers to extend agent capabilities. Auto-discover tools from registered servers, govern access through the policy engine.
Learn more about the MCP registry →
What Agents Can Access#
What a specific agent can access depends on its team's policies and the connections configured for the organization:
| Connection type | Access method | Credential handling |
|---|---|---|
| Kubernetes clusters | kubectl, K8s API | Short-lived JWT, impersonation |
| AWS accounts | AWS CLI, SDK | STS AssumeRole, session tags |
| GitHub | gh CLI, Git | Platform-managed tokens |
| Any HTTP system (Salesforce, Jira, ServiceNow, Datadog, PagerDuty, Zendesk, internal APIs) | HTTP/HTTPS via proxy | Header injection (Bearer, API key, Basic auth, custom) |
| MCP servers | MCP tools | Per-server authentication |
Supported agents and models#
Agents#
All agent types can use connections:
- Desktop AI tools: Claude Desktop, ChatGPT, Copilot, Cursor, Claude Code
- External agents: LangChain, CrewAI, Claude Agent SDK, AutoGen, any MCP-compatible framework
- Managed agents: agents created on the platform
Models#
Lens Agents integrates with Anthropic (Claude) and AWS Bedrock today. The platform is designed to be provider-agnostic; additional providers are added based on customer need. See Supported Models for current integrations and the extensibility model.
The model runs the agent's intelligence. Lens Agents provides the tools and connectivity. These are independent — you choose your model, Lens Agents handles access.
Shell tools#
Sandboxed environments include pre-installed CLI tools for automation:
aws, kubectl, gh, jq, yq, ripgrep, webfetch, git, curl, make, gcc, mise
Additional tools can be installed at runtime via mise (runtime version manager).
Related#
- Policies — controlling what agents can access
- Credential bindings — how credentials are injected
- Sandbox isolation — network isolation and proxy enforcement