mcp-server-claude-chat

An MCP (Model Context Protocol) server for the Anthropic Claude Messages API. Built in Rust, it exposes chat, vision, live web search, and model listing as MCP tools — so any MCP client (Claude Desktop, Claude Code, etc.) can consult Claude as a tool.

Communicates via stdio using JSON-RPC 2.0, like all the other MCP servers in this collection. Structurally it mirrors mcp-server-grok-chat / mcp-server-gpt-chat, adapted to Anthropic's API shape.

Subscription vs. API credits

This server uses the Anthropic API with an API key (API credits). That is the only supported, terms-compliant way to drive Claude from a third-party tool.

A Claude Pro/Max subscription (the kind that powers Claude Code and the Claude apps) is not usable here: that authentication is licensed for Anthropic's own first-party clients, and routing it through a custom server would violate Anthropic's Consumer Terms — the "$200 of Claude usage from another service" concern you raised. So: API key it is. (If you run an Anthropic-compatible gateway, point base_url at it — see Configuration.)

Tools

Tool	Description
`chat`	Send a message to Claude. Supports multi-turn history, a system prompt, model selection, and extended thinking via `thinking_budget`.
`chat_with_vision`	Analyse an image given an image URL and a text prompt.
`chat_with_search`	Chat with live web search — Claude searches server-side and answers with cited sources.
`list_models`	List available Claude models and their IDs (cached for 5 minutes).

Note: Anthropic has no embeddings endpoint (they recommend Voyage AI), so unlike the grok/gpt servers there is no embedding tool.

chat

Name	Type	Required	Description
`prompt`	string	yes	The user message to send
`system_prompt`	string	no	System prompt to set context/behaviour
`messages`	string	no	Conversation history as a JSON array of `{role, content}` objects (roles: `user`/`assistant` only — use `system_prompt` for system instructions)
`model`	string	no	Model ID (default: server default, `claude-opus-4-8`). Call `list_models`.
`temperature`	float	no	Sampling temperature (0.0–1.0). Ignored when `thinking_budget` is set.
`max_tokens`	integer	no	Max tokens to generate (defaults to the server default)
`thinking_budget`	integer	no	Enable extended thinking with this token budget (min 1024). Reasoning is returned in a `[thinking]` block; `max_tokens` is auto-raised above the budget.

chat_with_vision

Name	Type	Required	Description
`prompt`	string	yes	Text prompt describing what to analyse
`image_url`	string	yes	URL of the image (must be `http://` or `https://`)
`model`	string	no	Vision-capable model ID (default: server default)
`temperature`	float	no	Sampling temperature (0.0–1.0)
`max_tokens`	integer	no	Max tokens to generate

chat_with_search

Name	Type	Required	Description
`prompt`	string	yes	The user message
`system_prompt`	string	no	System prompt
`model`	string	no	Model ID (default: server default)
`temperature`	float	no	Sampling temperature (0.0–1.0)
`max_tokens`	integer	no	Max tokens to generate
`max_uses`	integer	no	Max web searches Claude may run for this request (default: 5)

Web search is billed by Anthropic at $10 per 1,000 searches on top of token costs, and an org admin must enable web search in the Claude Console.

list_models

No parameters. Results cached for 5 minutes.

Prerequisites

Rust (edition 2024)
An Anthropic API key from console.anthropic.com

Setup

bash

mkdir -p ~/.config/mcp-server-claude-chat
cp config.toml.example ~/.config/mcp-server-claude-chat/config.toml
# then edit it and set api_key

Config (~/.config/mcp-server-claude-chat/config.toml):

toml

api_key = "sk-ant-..."

# Optional:
# base_url = "https://api.anthropic.com/v1"   # or a compatible gateway
# default_model = "claude-opus-4-8"
# default_max_tokens = 4096

The server fails fast at startup if the config is missing or api_key is empty.

Build

bash

cargo build --release   # produces target/release/claude-chat
cargo build             # debug build
cargo run               # run in dev mode
RUST_LOG=debug cargo run
cargo test              # unit tests (response formatting, validation, message building)

MCP Configuration

Claude Desktop (~/.config/Claude/claude_desktop_config.json) or any MCP client:

json

{
  "mcpServers": {
    "claude-chat": {
      "command": "/media/codechap/4TB/develop/mcps/mcp-server-claude-chat/target/release/claude-chat"
    }
  }
}

Claude Code:

bash

claude mcp add claude-chat -- /media/codechap/4TB/develop/mcps/mcp-server-claude-chat/target/release/claude-chat

Project Structure

src/
  main.rs    - entry point, config loading, stdio transport setup
  server.rs  - MCP tool definitions (chat, chat_with_vision, chat_with_search, list_models)
  api.rs     - Anthropic HTTP client, Messages/Models types, response formatter
  params.rs  - tool parameter types with serde + JSON Schema derives
  config.rs  - TOML config loading