Inkeep MCP & RAG API

Nick Gomez

This week we launched a new API: inkeep-rag, which lets you query your docs, help center content, knowledge bases, and any other source ingested by Inkeep.

With this API, you can use your own content to power custom support agents, copilots, or custom workflows.

Let's see how to use it as part of an MCP Server, so we can connect it to Cursor or Anthropic's Claude Desktop App.

API Request

The inkeep-rag API takes in a query and outputs documents or chunks of content that can be passed directly to an LLM.

The API is exposed in the OpenAI chat completions format. The “query” is actually just a messages array, making it interoperable with the formats used by OpenAI, Anthropic, or other LLM providers. We can use any OpenAI compatible SDK to make the request.

python
Copy code
from openai import OpenAI

client = OpenAI(
    base_url=INKEEP_API_BASE_URL,
    api_key=INKEEP_API_KEY,
)

response = client.chat.completions.create(
    model="inkeep-rag",
    messages=[
        {"role": "user", "content": "How do I get started with Inkeep?"},
    ]
)

The messages array can contain full conversations, system prompts, and even images. The Inkeep RAG service automatically handles any summarization or intermediary steps for finding the most relevant content for the conversation.

API Response

The inkeep-rag API responds with content containing an array of documents, each in the format of:

jsx
Copy code
{
    "type": "document",
    "source": {
        "type": "text",
        "media_type": "text/plain",
        "data": "To get started with Inkeep..."
    },
    "title": "Getting Started",
    "context": "This is a docs page.",
    "record_type": "documentation", // inkeep specific field
    "url": "https://docs.inkeep.com/overview/ai-for-customers", // inkeep specific field
}

Using the content

This format extends Anthropic's Claude Citations, so can easily be used by Anthropic's Claude models and generally by any LLM or tool.

To call Claude directly:

python
Copy code
import anthropic

client = anthropic.Anthropic()

# query = "How to get started with Inkeep?"
documents_from_inkeep_rag_model = [...]

response = client.messages.create(
    model="claude-3-7-sonnet-20250219",
    max_tokens=1024,
    messages=[
        {
            "role": "user",
            "content": [
                *documents_from_inkeep_rag_model,
                {
                    "type": "text",
                    "text": "Write a blog post on how to get started"
                }
            ]
        }
    ]
)

Using it within an MCP Server

To use this RAG model with Cursor, Anthropic's Claude Desktop App, or other MCP compatible clients, use this template to get started.

You can customize the MCP Server with any business logic you like. As a basic example, we can create an MCP Tool that takes in a query and calls the Inkeep RAG API.

python
Copy code
import openai
from mcp.server.fastmcp import FastMCP
from pydantic import BaseModel

mcp = FastMCP("inkeep-mcp-server")

@mcp.tool(
    name="retrieve-product-docs",
    description="Retrieve product docs about your product using the Inkeep RAG API. 'query' should be framed as a conversational question about the product.",
)
async def retrieve_product_docs(query: str) -> InkeepRAGResponse:
    # make the call to Inkeep's RAG API
    return res

Getting started

As you can see - we lean on using formats and standards already common and established so it's easy to plug in Inkeep to any LLM application or use case.

To get started with our new RAG API, check out the open source repo to set up your own local MCP server powered by your own content.