Problem
If you happen to work with proprietary or private internal APIs, the LLM your coding assistant is leveraging very likely is not trained on what you need for these APIs.
If you want that boost of productivity AI based coding assistance provides, you're SOL.
Solution
Simple, MCP + RAG
Details
You can ingest your proprietary API documentation and give them semantic search capability through the magic of a vector store, an absolute essential for your coding assistant queries.
You'll add MCP to the mix to make your proprietary data vector store accessible to your coding assistant.
┌─────────────────┐ ┌─────────────────┐ ┌─────────────────┐
│ Cursor/VSCode │ │ MCP Server │ │ Vector Store │
│ │◄──►│ │◄──►│ (Chroma) │
│ AI Assistant │ │ │ │ API Docs RAG │
└─────────────────┘ └─────────────────┘ └─────────────────┘
|
┌─────────────────┐
│ Documentation │
│ Loader │
│ (Web + YAML) │
└─────────────────┘
Thoughts
All companies as well as 3rd party providers they partner with have proprietary API documentation which could be of use to application developers if this were available to them via their coding assistant.
Such coding assistant enhancements could greatly accelerate the developer's productivity and effectiveness.
Implementation of MCP and RAG in this manner should no longer be considered the domain of dedicated AI developers or teams.
Every developer needs the ability to tap into the company's knowledge bases to build better applications more quickly.
Open Source Project Idea
I created a Github repo to provide the building blocks with instructions intended to be extended by any application developer to get his/her feet wet with MCP + RAG to enhance their code-assistant capabilities with company KB data sources.
Let me know what you think of this idea. What I have created is not quite ready for prime time, but as an open source project perhaps community contributions could make it so.
https://github.com/ississippi/bring-your-own-rag
Top comments (0)