# claude-chat [Health: Active]

**Category:** 🗣️ Conversational AI  
**Repository:** https://github.com/codeChap/mcp-server-claude-chat  
**GitHub Stars:** 0  
**Views:** 3  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/claude-chat

## Description
Anthropic Claude Messages API MCP server — chat, vision, web search, extended thinking

## Tools
Capabilities this server exposes over MCP:

- **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).
- **prompt** — The user message to send
- **system_prompt** — System prompt to set context/behaviour
- **messages** — Conversation history as a JSON array of `{role, content}` objects (roles: `user`/`assistant` only — use `system_prompt` for system instructions)
- **model** — Model ID (default: server default, `claude-opus-4-8`). Call `list_models`.
- **temperature** — Sampling temperature (0.0–1.0). Ignored when `thinking_budget` is set.
- **max_tokens** — Max tokens to generate (defaults to the server default)
- **thinking_budget** — Enable extended thinking with this token budget (min 1024). Reasoning is returned in a `[thinking]` block; `max_tokens` is auto-raised above the budget.
- **image_url** — URL of the image (must be `http://` or `https://`)
- **max_uses** — Max web searches Claude may run for this request (default: 5)

## Claude Desktop Quick Installation
No confirmed install details yet, so no config is published. Follow the setup steps in the repository above.

## Documentation

## What claude-chat MCP server does

The claude-chat MCP server exposes Anthropic's Messages API as MCP tools for clients that communicate over stdio. It supports ordinary Claude conversations, image analysis, live web search, and model discovery. The implementation is written in Rust and sends requests to Anthropic using an API key funded with API credits.

The `chat` tool accepts a required prompt and can receive prior user and assistant messages as a JSON array. Optional parameters set a system prompt, model, temperature, output limit, and extended-thinking budget. When thinking is enabled, the response includes reasoning in a `[thinking]` block, and the server raises the output limit above the requested thinking budget.

## How it works

MCP clients start the compiled `claude-chat` executable and communicate with it through standard input and output. The process uses JSON-RPC 2.0 to expose four tools: `chat`, `chat_with_vision`, `chat_with_search`, and `list_models`.

Vision requests require both a text prompt and an HTTP or HTTPS image URL. Search requests let Claude perform live web searches on the server side and return answers with cited sources. The search tool accepts `max_uses`, which defaults to five searches for a request. Model results from `list_models` are cached for five minutes.

## Setup and configuration

The claude-chat MCP server requires Rust with edition 2024 support and an Anthropic API key from the Anthropic Console. Create `~/.config/mcp-server-claude-chat/config.toml` from the provided example, then set the required `api_key` value. Startup fails if the configuration file is absent or the key is empty.

Optional TOML settings include `base_url`, `default_model`, and `default_max_tokens`. `base_url` can point to an Anthropic-compatible gateway instead of the default Anthropic endpoint. Build the release executable with `cargo build --release`, then register its path in an MCP client's server configuration. Claude Desktop and Claude Code examples are provided in the project documentation.

## Tools and capabilities

- `chat` supports single-turn and multi-turn text requests, system prompts, model selection, temperature, maximum output tokens, and extended thinking.
- `chat_with_vision` sends a prompt and remote image URL to a vision-capable Claude model.
- `chat_with_search` enables Claude's live web search and returns cited results.
- `list_models` reports available Claude model IDs without parameters.

## Limitations and notes

The claude-chat MCP server requires Anthropic API credits; Claude Pro or Max subscriptions are not supported for this third-party server. Web search must be enabled by an organization administrator in the Claude Console, and Anthropic charges an additional $10 per 1,000 searches on top of token costs.

Conversation history for `chat` must contain only `user` and `assistant` roles. Put system instructions in `system_prompt` instead. The vision tool accepts image URLs only, not local file paths. Anthropic does not provide an embeddings endpoint in this server, so there is no embedding tool.

The project is licensed under MIT.

_Full upstream README: https://allmcps.com/mcp/claude-chat/readme_