Run Anthropic Claude models in Melis
The Claude provider implements the AI engine’s contract for Anthropic’s Messages API — selected automatically when a model’s company is Anthropic. Just add a Claude model, store the key, and point an agent at it.
// Use Claude: // 1. create a model whose company is Anthropic // 2. model id like 'claude-…', store the API key // 3. point an agent / instance at it // The engine routes here by company name — // every chat then goes through Claude.
Package
Role
Vendor
API
Message key
Tables
Overview
How Melis talks to Claude
MelisAIEngineModelClaudeService extends the engine contract and POSTs to Anthropic’s Messages API with the model’s API key (x-api-key) and the anthropic-version header. It builds Anthropic content blocks — text, image and document (PDF).
It parses tool_use blocks, dispatches each tool (MCP or built-in), feeds the tool_result back, and loops until stop_reason is end_turn. Anthropic prompt caching and a built-in web_fetch tool are enabled automatically.
Key features
A provider built for agentic Claude
The Messages API, content blocks, the tool-use loop and Anthropic’s caching and web tools — wired into the engine’s contract.
Messages API
Talks to Anthropic’s Messages endpoint with the model’s API key.
Content blocks
Text, image and document (PDF) blocks built per the contract.
Tool use
Parses tool_use, dispatches each tool, feeds tool_result back, loops until end_turn.
Prompt caching
Enables Anthropic prompt caching to cut cost on repeated context.
Built-in web_fetch
Appends Anthropic’s web_fetch tool automatically.
Tokens & files
Reads usage (input/output + cache); embeds files (base64 / extracted text) by default.
See it in action
Engine tools mapped to Anthropic shape
The provider implements the engine contract and forwards your tools as Anthropic input_schema — then loops on tool_use until the turn ends.
class MelisAIEngineModelClaudeService extends MelisAIEngineModelService { public function getMessageKey(): string { return 'messages'; } // addToolsToPayload(): Anthropic input_schema, + web_fetch // sendCustomAI(): parse tool_use → dispatch → loop to end_turn }
Engine tools → Anthropic shape
Engine tools already use Anthropic’s input_schema, so the provider sanitises and forwards them; a tool_use call is correlated to its result by id.
- Message key: messages
- Roles: user / assistant
- tool_use_id correlates call ↔ result
For developers
How it's wired
Auth & version
x-api-key (the model’s key) plus the anthropic-version header; a long timeout for tool loops.
Stateless
No tables — conversation state lives in the engine.
DB-only key
The API key comes from the model’s platform key, not an environment variable.
// Selected when the model's company contains 'Anthropic'. // Returns tool results as {type:'tool_result', tool_use_id, content} // inside a user message, then continues until stop_reason==='end_turn'.
Part of Melis Platform
Explore the rest of the platform
The Claude provider plugs into the AI engine — here are the modules around it.
Add Claude to your platform
See agentic AI running on Anthropic Claude.
