Extending Diogenes¶

Diogenes can be extended at four layers: skill Markdown, MCP servers, built-in Python tools, and HTTP integrations. Each layer has a different registration and execution path.

Choose an extension path¶

Execution map¶

flowchart TD
    A[model request] --> B[tool index]
    B --> C[OpenAI function schemas]
    C --> D[tool parsing]
    D --> E[policy and approval]
    E --> F{qualified MCP name?}
    F -- yes --> G[McpManager.call_tool]
    F -- no --> H[TOOL_HANDLERS]
    H --> I[in-process handler]
    G --> J[paired tool result]
    I --> J

src/tool_index.py selects a small schema set for each prompt. src/tool_schemas.py supplies OpenAI-compatible schemas. src/tool_parsing.py normalizes native calls and fenced call formats. src/tool_execution.py applies policy and invokes the selected handler.

Development pages¶

• Skills and instructions covers the skill file format, indexing, imports, tests, and audits.
• MCP and built-in tools covers server registration and a complete in-process tool addition.
• API, CLI, and companion covers bearer tokens, chat requests, pairing, and command dispatch.
• Integrations and webhooks covers HTTP service records, api_call, and event delivery.

Minimum test set¶

An extension should cover:

1. valid registration;
2. malformed configuration;
3. unavailable dependency;
4. user and role filtering;
5. approval classification;
6. tool result pairing;
7. redaction of credentials;
8. browser rendering of errors.