AI · Integrations
Model Context Protocol on VegaDūta
VegaDūta speaks Model Context Protocol in both directions. As a client, it connects your agents to any MCP server — from the official registry or your own — over STDIO, SSE, or streamable HTTP. As a server, it exposes five curated tools over your own account: list your agents, call an agent, list your workflows, call a workflow, and search your knowledge base. Any MCP client — Claude, an IDE, another platform — connects with a scoped API key.
You browse the official MCP registry (registry.modelcontextprotocol.io) directly from the console with cached search, or register a custom server dynamically with its command, arguments, and encrypted environment credentials. Registrations are editable and removable after the fact, not write-once. Home Assistant and Atlassian (Jira/Confluence) ship as first-party connectors, and Atlassian connects in one click over OAuth with no API token to copy.
Connect MCP servers to your agents
Every transport in the MCP spec is supported: STDIO for local processes, SSE, and streamable HTTP for remote servers. The platform tracks connection liveness and self-heals STDIO servers that die, so a crashed local process does not silently strand your agent without its tools.
- Browse and search the official MCP registry from the console, backed by a cached catalog
- Register custom servers with command/args and encrypted env credentials
- One-click OAuth connectors where the vendor supports it — Atlassian's Remote MCP server needs no token and no app registration
- Edit or delete a registration later — change a URL, rotate a credential, or remove a server without starting over
- Connection liveness tracking with self-healing restarts for STDIO servers
- Per-plan connector limits (see /billing for plan details)
Security around third-party servers
MCP servers are third-party code, so VegaDūta wraps them in layered controls rather than trusting them by default. URLs pass an SSRF guard, STDIO commands run against an allowlist inside a sandbox, and tool arguments and outputs are sanitized before they reach a model or leave the platform.
- SSRF URL guard and STDIO command allowlist with sandboxed execution
- Tool-argument and tool-output sanitization plus credential redaction
- Admin review queue for new server registrations
- Per-tool-call human approval gates on sensitive actions
Expose VegaDūta as an MCP server
The same protocol runs outward. Connect an MCP client to VegaDūta with a scoped API key and it sees exactly five tools — a deliberately curated set, not your whole internal toolbox re-exported. Rate limits apply per key, and you decide which capabilities each external client is granted.
- list_my_agents — the agents this key is allowed to see
- call_agent — run one of them and get its answer back
- list_my_workflows and call_workflow — the same for workflows
- search_knowledge — retrieval over your own uploaded documents
Enterprise systems: what is and is not built in
Home Assistant and Atlassian (Jira/Confluence) are the first-party connectors. SAP and similar enterprise systems do not have built-in connectors — but if an MCP server exists for them, you can register it from the registry or as a custom server and your agents get its tools the same way. That is the point of standardizing on MCP instead of hand-writing one integration per system.
Frequently asked questions
What is the Model Context Protocol (MCP)?
MCP is an open protocol that lets AI applications connect to external tools and data sources through a standard interface. Instead of writing a bespoke integration for every system, a client speaks MCP to any compliant server. VegaDūta implements the protocol as both a client and a server.
How do I connect an MCP server to VegaDūta?
Browse the official MCP registry from the console and add a server with cached search, or register a custom one by supplying its command and arguments (STDIO) or URL (SSE / streamable HTTP). Credentials go into encrypted environment variables. Where a vendor supports OAuth — Atlassian, for example — connecting is a single click with no token to copy. Registrations can pass through an admin review queue before agents use them, and can be edited or deleted afterwards.
Can Claude or my IDE use my VegaDūta agents as tools?
Yes. Connect any MCP client with a scoped API key and it gets five tools: list_my_agents, call_agent, list_my_workflows, call_workflow, and search_knowledge. That curated set is intentional — the outward server exposes your account's agents, workflows and knowledge, not the internal toolbox those agents use. Rate limits apply per key.
Can I connect Jira and Confluence to my agents?
Yes, in one click. VegaDūta connects to Atlassian's official Remote MCP server over OAuth 2.1, using a single platform-wide client registered through Dynamic Client Registration — so there is no API token to generate and no Atlassian app for you to register first.
Is it safe to run third-party MCP servers?
VegaDūta assumes they are untrusted and layers controls around them: an SSRF guard on URLs, a command allowlist and sandbox for STDIO servers, sanitization of tool arguments and outputs, and credential redaction. For risky actions you can require a human to approve each tool call.
See it working in two minutes
The sandbox provisions a real tenant — describe an agent in one sentence and test it, no account, no card. Or browse ~90 industry workflow recipes to see what teams build.
