Consult7

What is Consult7 about?

Consult7 provides a Model Context Protocol (MCP) server that lets AI agents send a set of files (paths with optional wildcards) together with a natural‑language query to a language model that offers a massive context window. The model processes the full assembled context and returns the answer directly to the agent.

How to use Consult7?

1. Install – No manual install is required; uvx downloads and runs the tool in an isolated environment.
2. Add the server – For Claude Code run:

   # OpenRouter
   claude mcp add -s user consult7 uvx -- consult7 openrouter <your-api-key>
   
   # Google AI
   claude mcp add -s user consult7 uvx -- consult7 google <your-api-key>
   
   # OpenAI
   claude mcp add -s user consult7 uvx -- consult7 openai <your-api-key>
   

   For Claude Desktop edit the mcpServers section of the config file accordingly.
3. Invoke – Use the MCP from the agent, specifying the model (e.g., gemini-2.5-flash|thinking) and a list of absolute file paths/wildcards.
4. Optional testing – Verify the API key with uvx consult7 <provider> <api-key> --test.

Key Features

• Supports three providers: OpenRouter, Google AI, OpenAI.
• Works with any model that offers a large context window (up to >1 M tokens).
• Simple file‑specification rules (absolute paths, wildcards only in filenames, automatic ignores for common noise files).
• "Thinking" mode for deeper reasoning on supported models.
• Size limits: ≤1 MB per file, ≤4 MB total, optimized for ~1 M token windows.
• No permanent installation; uvx handles isolated execution.

Use Cases

• Summarize an entire codebase – Feed all source files and ask for architecture overview.
• Locate specific implementations – Search for a method across mixed Python/JS projects.
• Analyze test coverage – List test files and highlight uncovered components.
• Security review – Step‑by‑step analysis of authentication flows with reasoning mode.
• Document repository inspection – Query large collections of markdown, PDFs, or mixed formats.

FAQ

Q: Do I need to install anything locally? A: No. uvx automatically downloads the Consult7 binary in a sandboxed environment.

Q: Which models can I use? A: Provider‑specific examples are listed in the README – e.g., gemini-2.5-pro for Google, anthropic/claude-opus-4.1|thinking for OpenRouter, gpt-5|400k for OpenAI.

Q: How are large files handled? A: Files larger than 1 MB are ignored; the total payload is capped at 4 MB to stay within typical context limits.

Q: Can I customize reasoning tokens? A: Yes, using the |thinking=30000 suffix, though the default is usually sufficient.

Q: How do I remove the server? A: Run claude mcp remove consult7 -s user.

Consult7 MCP Server

Consult7 is a Model Context Protocol (MCP) server that enables AI agents to consult large context window models for analyzing extensive file collections - entire codebases, document repositories, or mixed content that exceed the current agent's context limits. Supports providers Openrouter, OpenAI, and Google.

Why Consult7?

When working with AI agents that have limited context windows (like Claude with 200K tokens), Consult7 allows them to leverage models with massive context windows to analyze large codebases or document collections that would otherwise be impossible to process in a single query.

"For Claude Code users, Consult7 is a game changer."

How it works

Consult7 collects files from the specific paths you provide (with optional wildcards in filenames), assembles them into a single context, and sends them to a large context window model along with your query. The result is directly fed back to the agent you are working with.

Example Use Cases

Summarize an entire codebase

• Files: ["/Users/john/project/src/*.py", "/Users/john/project/lib/*.py"]
• Query: "Summarize the architecture and main components of this Python project"
• Model: "gemini-2.5-flash"

Find specific method definitions

• Files: ["/Users/john/backend/src/*.py", "/Users/john/backend/auth/*.js"]
• Query: "Find the implementation of the authenticate_user method and explain how it handles password verification"
• Model: "gemini-2.5-pro"

Analyze test coverage

• Files: ["/Users/john/project/tests/*_test.py", "/Users/john/project/src/*.py"]
• Query: "List all the test files and identify which components lack test coverage"
• Model: "gemini-2.5-flash"

Complex analysis with thinking mode

• Files: ["/Users/john/webapp/src/*.py", "/Users/john/webapp/auth/*.py", "/Users/john/webapp/api/*.js"]
• Query: "Analyze the authentication flow across this codebase. Think step by step about security vulnerabilities and suggest improvements"
• Model: "gemini-2.5-flash|thinking"

Installation

Claude Code

Simply run:

# OpenRouter
claude mcp add -s user consult7 uvx -- consult7 openrouter your-api-key

# Google AI
claude mcp add -s user consult7 uvx -- consult7 google your-api-key

# OpenAI
claude mcp add -s user consult7 uvx -- consult7 openai your-api-key

Claude Desktop

Add to your Claude Desktop configuration file:

{
  "mcpServers": {
    "consult7": {
      "type": "stdio",
      "command": "uvx",
      "args": ["consult7", "openrouter", "your-api-key"]
    }
  }
}

Replace openrouter with your provider choice (google or openai) and your-api-key with your actual API key.

No installation required - uvx automatically downloads and runs consult7 in an isolated environment.

Command Line Options

uvx consult7 <provider> <api-key> [--test]

• <provider>: Required. Choose from openrouter, google, or openai
• <api-key>: Required. Your API key for the chosen provider
• --test: Optional. Test the API connection

The model is specified when calling the tool, not at startup. The server shows example models for your provider on startup.

Model Examples

Google

Standard models:

• "gemini-2.5-flash" - Fast model
• "gemini-2.5-flash-lite" - Ultra fast lite model
• "gemini-2.5-pro" - Intelligent model
• "gemini-2.0-flash-exp" - Experimental model

With thinking mode (add |thinking suffix):

• "gemini-2.5-flash|thinking" - Fast with deep reasoning
• "gemini-2.5-flash-lite|thinking" - Ultra fast with deep reasoning
• "gemini-2.5-pro|thinking" - Intelligent with deep reasoning

OpenRouter

Standard models:

• "google/gemini-2.5-pro" - Intelligent, 1M context
• "google/gemini-2.5-flash" - Fast, 1M context
• "google/gemini-2.5-flash-lite" - Ultra fast, 1M context
• "anthropic/claude-sonnet-4" - Claude Sonnet, 200k context
• "anthropic/claude-opus-4.1" - Claude Opus 4.1, 200k context
• "openai/gpt-5" - GPT-5, 400k context
• "openai/gpt-4.1" - GPT-4.1, 1M+ context

With reasoning mode (add |thinking suffix):

• "anthropic/claude-sonnet-4|thinking" - Claude with 31,999 reasoning tokens
• "anthropic/claude-opus-4.1|thinking" - Opus 4.1 with reasoning
• "google/gemini-2.5-flash-lite|thinking" - Ultra fast with reasoning
• "openai/gpt-5|thinking" - GPT-5 with reasoning
• "openai/gpt-4.1|thinking" - GPT-4.1 with reasoning effort=high

OpenAI

Standard models (include context length):

• "gpt-5|400k" - GPT-5, 400k context
• "gpt-5-mini|400k" - GPT-5 Mini, faster
• "gpt-5-nano|400k" - GPT-5 Nano, ultra fast
• "gpt-4.1-2025-04-14|1047576" - 1M+ context, very fast
• "gpt-4.1-nano-2025-04-14|1047576" - 1M+ context, ultra fast
• "o3-2025-04-16|200k" - Advanced reasoning model
• "o4-mini-2025-04-16|200k" - Fast reasoning model

O-series models with |thinking marker:

• "o1-mini|128k|thinking" - Mini reasoning with |thinking marker
• "o3-2025-04-16|200k|thinking" - Advanced reasoning with |thinking marker

Note: For OpenAI, |thinking is only supported on o-series models and serves as an informational marker. The models use reasoning tokens automatically.

Advanced: You can specify custom thinking tokens with |thinking=30000 but this is rarely needed.

File Specification Rules

When using the consultation tool, you provide a list of file paths with these rules:

1. All paths must be absolute (start with /)

   • ✅ Good: /Users/john/project/src/*.py
   • ❌ Bad: src/*.py or ./src/*.py

2. Wildcards (*) only allowed in filenames, not in directory paths

   • ✅ Good: /Users/john/project/*.py
   • ❌ Bad: /Users/*/project/*.py or /Users/john/**/*.py

3. Must specify extension when using wildcards

   • ✅ Good: /Users/john/project/*.py
   • ❌ Bad: /Users/john/project/*

4. Mix specific files and patterns freely

   • ✅ Good: ["/path/src/*.py", "/path/README.md", "/path/tests/*_test.py"]

5. Common patterns:

   • All Python files in a directory: /path/to/dir/*.py
   • Test files: /path/to/tests/*_test.py or /path/to/tests/test_*.py
   • Multiple extensions: Use multiple patterns like ["/path/*.js", "/path/*.ts"]

The tool automatically ignores: __pycache__, .env, secrets.py, .DS_Store, .git, node_modules

Size limits: 1MB per file, 4MB total (optimized for ~1M token context windows)

Testing

# Test OpenRouter
uvx consult7 openrouter sk-or-v1-... --test

# Test Google AI
uvx consult7 google AIza... --test

# Test OpenAI
uvx consult7 openai sk-proj-... --test

Uninstalling

To remove consult7 from Claude Code (or before reinstalling):

claude mcp remove consult7 -s user