🚀 EDDI 6.5.0

19 LLM providers, knowledge bases that crawl your site and read your files, and conversations that follow compatible agent versionsWhat's new

Managed Cloud Get Started
All integrations
Agent protocolsmcpcalls

MCP Client

Give EDDI agents the tools of external MCP servers. A server's tools become the agent's tools, behind the same approval gates and audit trail as every other tool.

ConfigurationDocs ↗

What you get

  • Tools from any MCP server that speaks Streamable HTTP
  • Name allow and deny lists to keep the tool list small
  • Optional resource listing and reading
  • A circuit breaker that stops calling a failing server

How it works in EDDI

  1. Create an mcpcalls configuration with the server URL, the transport http and an API key as a ${vault:...} reference.
  2. Reference it as a workflow step (eddi://ai.labs.mcpcalls).
  3. The agent's LLM task picks up the server's tools automatically (enableMcpCallTools defaults to true).
  4. Add the tools to toolApprovals so consequential calls wait for a human.
  5. For a stdio-only server, run it in its own container behind an HTTP bridge and point EDDI at the bridge.

Configuration

Taken from the EDDI documentation. Placeholders such as your-agent-id are yours to replace.

mcpcalls configurationAn mcpcalls configuration. The token is a vault reference.
{
  "mcpServerUrl": "https://mcp.example.com/mcp",
  "name": "example",
  "transport": "http",
  "apiKey": "${vault:example-mcp-token}",
  "timeoutMs": 30000
}
agent configuration (hitlConfig)Gate every MCP tool except the read-only ones.
{
  "hitlConfig": {
    "toolApprovals": {
      "requireApproval": [
        "mcp:*"
      ],
      "exempt": [
        "mcp:read_*"
      ],
      "approvalTimeout": "PT30M",
      "timeoutPolicy": "AUTO_REJECT"
    }
  }
}

Governance

  • MCP tool calls take part in the same human-in-the-loop approval gate as every other tool source.
  • Tool descriptions are redacted for directive-shaped text and length-capped. Tool results are wrapped with provenance and, by default, directive-shaped content is redacted.
  • A tool whitelist is context management, not a security boundary. The boundary is the credential you give the connection plus the approval gate.
  • Every gated decision is written to the audit ledger.

Good to know

  • EDDI implements one MCP transport, Streamable HTTP. It does not spawn stdio servers itself, because a configuration-editable command would be arbitrary code execution as the EDDI process user. Use a bridge sidecar.
  • A circuit breaker opens after 3 failures in 60 seconds, so a missing tool usually means the server is failing.
Get startedDocs ↗

Related integrations

Agent protocolsMCP ServerControl EDDI from Claude Desktop, Cursor, VS Code and other MCP clients. EDDI exposes 84 MCP tools over Streamable HTTP for conversations and administration.Agent protocolsA2AExpose EDDI agents to other agents through Agent Cards and JSON-RPC, and use remote A2A agents as tools inside your own agents.Identity and securitySecrets VaultA built-in, envelope-encrypted secrets vault. Configuration references secrets as ${vault:keyName}, and the value never lands in agent configs, exports or logs.

Configuration and behavior on this page follow the EDDI documentation, checked on 2026-10-14. Field names and defaults can change between releases, so confirm them in the docs for your version.