MCP Integration Guide

Connect Claude, Gemini, and Codex to Monoshiri AI

Learn how to connect MCP-compatible AI clients directly to your internal knowledge base.

Wouldn't it be great if you could ask questions about your company's internal knowledge directly from the AI tools you already use?

Monoshiri AI supports MCP (Model Context Protocol), allowing you to connect directly to Monoshiri AI's knowledge base from any MCP-compatible AI client - including Claude Desktop, Claude.ai, Gemini CLI, Codex, Dify, and Cursor.

What You'll Learn

What MCP (Model Context Protocol) is
The four MCP tools provided by Monoshiri AI
How to connect from Claude Desktop / Claude.ai / Claude Code / Gemini CLI / Codex / Dify / Cursor
Security and access control

What Is MCP?

MCP (Model Context Protocol) is an open standard protocol that allows AI assistants to connect to external data sources and tools. Proposed by Anthropic, MCP is now supported by major AI tools including Claude, Gemini CLI, and Codex as of 2026.

With MCP, you can simply ask something like "Look up our PTO policy" in your regular AI chat, and it will retrieve the relevant information from Monoshiri AI's knowledge base to generate an answer. No need to switch between AI tools - your workflow stays uninterrupted.

Four MCP Tools Provided by Monoshiri AI

Monoshiri AI's MCP server provides the following four tools (operations).

Tool Name	What It Does
`list_folders`	Retrieve a list of folders you have access to
`list_documents`	Retrieve a list of documents in a specified folder
`search`	Generate an AI answer (with sources) from a specified folder's knowledge
`get_document`	Retrieve detailed information and content of a document

The search tool is the most important. When you submit a natural language query like "How do I submit an expense report?", the AI navigates the folder's documents to generate an answer with cited sources. folder_id is optional — if multiple folders exist, the response returns a selection list so you (or the AI) can pick the most relevant one and call search again.

Prerequisites

You have a Monoshiri AI account
Your team admin has enabled the MCP integration
MCP connections are read-only (mcp:read scope). You cannot add or delete documents through MCP

Connection Setup

The Monoshiri AI MCP server URL is:

https://api.monoshiri.ai/mcp

Authentication is done via OAuth 2.1 (recommended) or API key. When you connect using OAuth from any client, your browser will open to the Monoshiri AI login screen - simply authorize the connection and you're all set.

Connecting from each client

Claude Desktop

Anthropic

Claude Desktop is Anthropic's desktop application.

Setup steps

Open the following config file (create it if it doesn't exist)

macOS:~/Library/Application Support/Claude/claude_desktop_config.json
Windows:%APPDATA%\Claude\claude_desktop_config.json

Add the following:

json

{
  "mcpServers": {
    "monoshiri": {
      "url": "https://api.monoshiri.ai/mcp",
      "transport": "streamable-http"
    }
  }
}

Restart Claude Desktop. Your browser will automatically open to the Monoshiri AI OAuth authorization screen.

Click "Allow" to complete the connection.

Once connected, simply ask Claude something like "Tell me about our company's [topic]" and it will generate answers from Monoshiri AI's knowledge base.

Claude.ai

Web

You can also connect from the web version of Claude.ai as a custom connector.

Setup steps

Go to Settings > Connectors > Add custom connector

Enter https://api.monoshiri.ai/mcp as the Server URL

Click Connect - the OAuth authorization screen will open

Click "Allow" to complete the connection

Since everything runs in the browser, no app installation is required.

Claude Code

CLI

Claude Code is a terminal-based AI coding agent.

Setup steps

Add the following to your Claude Code config file (~/.claude/settings.json or .claude/settings.json in your project root).

json

{
  "mcpServers": {
    "monoshiri": {
      "url": "https://api.monoshiri.ai/mcp",
      "transport": "streamable-http"
    }
  }
}

Or add it directly from the CLI:

bash

claude mcp add monoshiri --transport streamable-http --url https://api.monoshiri.ai/mcp

During development, you can ask questions like "Check the spec for this feature" or "Search for past incident reports" directly from your terminal.

Gemini CLI

Google

Gemini CLI is Google's terminal-based AI agent, which supports MCP servers.

Setup steps

Add the following to your Gemini CLI config file (~/.gemini/settings.json).

json

{
  "mcpServers": {
    "monoshiri": {
      "url": "https://api.monoshiri.ai/mcp",
      "transport": "streamable-http"
    }
  }
}

Reference: Gemini CLI MCP documentation https://geminicli.com/docs/tools/mcp-server/

Codex CLI

OpenAI

Codex CLI is OpenAI's terminal-based AI agent. It supports MCP and can connect to Streamable HTTP servers.

Setup steps

Add it from the CLI:

bash

codex mcp add monoshiri --transport streamable-http --url https://api.monoshiri.ai/mcp

Or add it to your config file (~/.codex/config.toml or .codex/config.toml in your project root):

toml

[mcp_servers.monoshiri]
type = "url"
url = "https://api.monoshiri.ai/mcp"
transport = "streamable-http"

Reference: Codex MCP documentation https://developers.openai.com/codex/mcp

Dify

Dify is an open-source AI application development platform. It natively supports MCP since v1.6.0, allowing you to use MCP servers as tools.

Setup steps

In the Dify dashboard, go to Tools > MCP

Click Add Server

Enter https://api.monoshiri.ai/mcp as the Server URL

Configure authentication (use an API key)

Once saved, Monoshiri AI's search tools will be available in your Dify workflows and agents

By connecting Monoshiri AI through Dify, you can build custom AI agents powered by your internal knowledge - all without writing code.

Cursor

IDE

Cursor is an AI-powered code editor that supports MCP servers.

Setup steps

Create .cursor/mcp.json in your project root, or add the following to your user-level config at ~/.cursor/mcp.json:

json

{
  "mcpServers": {
    "monoshiri": {
      "url": "https://api.monoshiri.ai/mcp",
      "transport": "streamable-http"
    }
  }
}

While coding in Cursor, you can ask questions like "Check the requirements in this spec document" or "Search our internal API design guidelines" directly within the editor.

Verifying Your Connection: MCP Inspector

If your connection isn't working, you can troubleshoot using MCP Inspector.

bash

npx @modelcontextprotocol/inspector

After launching:

Transport:Select "Streamable HTTP"
URL:Enter https://api.monoshiri.ai/mcp
Connect:Click the button to start the OAuth flow

Once connected successfully, you'll see the four tools: list_folders, list_documents, search, and get_document.

Security and Access Control

Authentication

Method	Use Case	Expiration
OAuth 2.1 (recommended)	Claude Desktop / Claude.ai / Cursor, etc.	Access token: 15 min, Refresh token: up to 60 days
API Key	CI/scripts / Dify, etc.	Valid until revoked by an admin

Access Scope

MCP is read-only (mcp:read scope)
Only folders the user has access to are included in search results
Folder-level access controls remain enforced through MCP
You cannot add, delete, or modify documents via MCP

Admin Controls

Team admins can enable or disable the MCP integration at any time
Disabling it immediately terminates all member MCP connections
Only admins can issue or revoke API keys

Use Cases

Scenario	How to Use
Check specs during development	Ask Claude Code / Cursor / Codex: "Check the spec for this feature"
Look up info during meetings	Ask Claude.ai: "What was decided in the last meeting notes?"
Build custom AI agents	Create a Dify workflow that incorporates Monoshiri AI search
Everyday work questions	Ask Claude Desktop: "How do I submit an expense report?"
Code review	Ask Cursor: "Does this code follow our internal coding guidelines?"

Summary

With Monoshiri AI's MCP integration, you can ask questions about your internal knowledge directly from the AI tools you already use.

Claude Desktop / Claude.ai / Claude Code: Just add three lines to a JSON config file to connect
Gemini CLI: Connect from Google's terminal-based AI
Codex CLI: Connect from OpenAI's coding agent
Dify: Build custom AI agents without writing code
Cursor: Look up specs while coding in an AI-powered editor
Security: OAuth 2.1 compliant, read-only access, folder-level access control

Access your internal knowledge without ever leaving your current workflow. That's the real value of MCP integration.

You can also find MCP connection instructions on the account settings page in the Monoshiri AI dashboard.

Try Monoshiri AI for free

Just upload your documents and start asking AI. Try our free plan with unlimited users.

Get Started Free

No credit card required / Start in 1 minute