EDDI

What EDDI does

The EDDI MCP server is middleware for building and operating conversational AI applications from configuration. It coordinates users, AI agents, and business systems through routing, persistent memory, and API orchestration. Agent creation is handled through the Manager dashboard, including a form-based wizard and platform-operator workflows, so basic agent setup does not require writing application code.

EDDI belongs to the Conversational AI category and is designed for deployments that need more than a single chat model. Its documented protocol and platform support includes MCP, A2A, the OpenAI Chat Completions API, Slack, OpenAPI, and OAuth 2.0. The project also describes support for multiple LLM providers, RAG, persistent memory, and compliance-oriented requirements including the EU AI Act, GDPR, and HIPAA.

How it works

Configuration defines the agents and the way requests move through the system. EDDI can route conversations among agents, retain information through persistent memory, and connect agent workflows with external business systems through APIs and supported protocols. The platform is built on Quarkus and Java 25 and is distributed as a Docker image.

A deployment exposes one main port by default. The root address opens a chooser for the Manager at /manage and the Workforce workspace at /workforce; the standalone chat interface is available at /chat. Docker Compose files provide optional deployment components, including PostgreSQL, MongoDB, Keycloak authentication, Prometheus and Grafana monitoring, NATS, Ollama, Chroma, and an MCP bridge sidecar for stdio-only MCP servers.

Setup and configuration

The quickest installation path uses the supplied Linux, macOS, WSL2, or Windows installer and requires Docker. The installer starts EDDI with a selected database through Docker Compose and generates a vault encryption key for secret management. Supported installer options include default installation, PostgreSQL, authentication, monitoring, local image builds, release pinning, and dry-run inspection.

Manual Docker Compose operation is also supported. The repository includes separate compose files for database choices and optional services; the PostgreSQL stack is standalone rather than an overlay on the default MongoDB stack. Local model deployments can use Ollama, with an additional NVIDIA overlay for GPU access. An eddi update wrapper can pull and restart the latest image, while manual Docker commands are available when the wrapper is not present.

Tools and capabilities

The EDDI MCP server supports these documented capabilities:

• Define and manage AI agents through configuration and the Manager dashboard.
• Route conversations across multiple agents.
• Maintain persistent conversational memory.
• Orchestrate APIs and business-system interactions.
• Connect workflows through MCP and A2A.
• Expose or consume functionality through OpenAI Chat Completions API, Slack, OpenAPI, and OAuth 2.0 support.
• Run chat and group-conversation experiences through the Chat UI and Workforce workspace.
• Add databases, authentication, monitoring, vector storage, local models, or an MCP bridge through Compose configurations.

Limitations and notes

The supplied material does not identify individual MCP tool names, detailed JSON schemas, supported LLM provider names, or client-specific setup instructions. It also does not state that EDDI is compatible with a particular desktop MCP client. Review the project documentation before selecting a provider, configuring production authentication, or planning an upgrade; the README calls out release-specific upgrade steps and defaults that may fail closed.

EDDI is licensed under Apache 2.0. The standard quick start assumes Docker, and database, authentication, monitoring, and local-model features require the corresponding Compose configuration rather than being implied by the base deployment.