Google Developer Knowledge API: What Changes for You

7 min read

Came here from Instagram, LinkedIn, X or YouTube? Here is the full source list you came for. Everything below is built from these, nothing else.

The announcement

• Supercharge your development with the Google Developer Knowledge API ecosystem, Google Developers Blog, Oct 7 2026

On October 7, 2026, Google published a post on its Developers Blog introducing the Developer Knowledge API ecosystem. The API is described as Google's "official, programmatic source of truth" for developer documentation about Google Cloud, Firebase, Android, and more. It returns that documentation as fresh Markdown, built for AI agents, IDE extensions and automated workflows.

It ships with four ways in: a gcloud CLI surface, an official agent skill that talks to a Developer Knowledge MCP server, client libraries in 7 languages, and an interactive page in the Google APIs Explorer. The post was written by Christina Gonzalez, a Technical Writer in Software Engineering at Google.

Key takeaways

• Google now offers an official API for its own developer docs, covering Google Cloud, Firebase, Android and more, served as Markdown.
• It does semantic and keyword search, document chunking and grounded Q&A, and is indexed frequently so agents see updates with minimal lag.
• One command, npx skills add google/skills --skill retrieving-developer-knowledge, gives your coding assistant the skill. Google names Claude Code, Cursor, GitHub Copilot and Google Antigravity as compatible through MCP.
• The gcloud surface comes pre-installed in Cloud Shell and has 3 commands: answer-query, documents search-chunks and documents describe.
• Client libraries ship in 7 languages, and BatchGetDocuments pulls up to 20 documents in one call.

What is the Google Developer Knowledge API?

It is an API whose only job is to hand you Google's documentation in a shape a program can use.

Today, if your agent needs to know how Firestore transactions work, it has two bad options. It can answer from its training data, which stops at a cutoff date. Or it can scrape a docs page, which is fragile and fills the context window with navigation, footers and HTML.

Google's pitch is that this API replaces both. You send a query, you get back Markdown chunks from the official docs, or a full page, or a grounded answer. The post says it is "indexed frequently to ensure predictable retrieval, fresh documentation, and minimal lag between upstream updates and agent access."

The four core operations, as named in the client libraries, are:

Operation	What it does
SearchDocumentChunks	Searches snippets and chunks across Google documentation
GetDocument	Fetches one full document
BatchGetDocuments	Fetches up to 20 documents in a single call
AnswerQuery	Returns a grounded answer to a question

Why does this matter to a developer?

Because the most annoying class of AI coding mistake is the confident, outdated one.

You ask your assistant to set up a Google Cloud resource. It writes code that looks right, uses a flag or method that changed after its training cutoff, and you lose twenty minutes finding out. The model was not wrong about how things worked. It was wrong about when.

The post is direct about this. It tells you to "stop relying on brittle web-scraping or wrestling with stale LLM training cutoffs." The agent skill is described as "reducing hallucinations from stale LLM training cutoffs" by giving assistants access to current Google docs.

There is also a token angle, and this is the part I care about most. Most of Art of Code gets written with an AI coding assistant in the loop, and every page of docs it pulls in costs tokens. The skill uses a multi-step retrieval workflow. It searches chunks with searchDocumentChunks first, and only then fetches complete Markdown pages with documents.get, or skips both and asks answerQuery for a grounded answer. Small pieces first, full pages only when needed.

That is how you would design it by hand if you were watching your context size. Now it is the default.

What changes in practice?

It depends on where you work. There are four surfaces.

In your terminal

If you already have the Google Cloud SDK, you already have this. The post says the gcloud surface "comes pre-installed in Cloud Shell" and works across standard SDK installs on Linux, macOS and Windows "without requiring extra plugins or configuration."

These are the examples from the announcement:

# Get grounded answers
gcloud developer-knowledge answer-query --query="How do I create a BigQuery dataset?"

# Search document chunks
gcloud developer-knowledge documents search-chunks --query="Firestore transactions"

# Retrieve metadata and content for a specific document
gcloud developer-knowledge documents describe "documents/docs.cloud.google.com/storage/docs/creating-buckets"

# Troubleshoot an error trace by piping it directly into the answer-query command
gcloud developer-knowledge answer-query --query="$(cat error.txt)"

The last one is the one I would actually use. You save a stack trace to a file, pipe it in, and get an answer grounded in the official docs rather than a forum thread from three versions ago.

In your AI coding assistant

This is the biggest shift. Install the skill in your project:

npx skills add google/skills --skill retrieving-developer-knowledge

The skill tells your assistant how to use the Developer Knowledge MCP server, and to fall back to the REST API with curl when it needs to. Google lists Google Antigravity, Claude Code, Cursor, GitHub Copilot and custom agentic frameworks as compatible through standard MCP.

To connect the MCP server in your IDE, you authenticate with Application Default Credentials or an API key. The announcement points to a separate "Connect to the Developer Knowledge MCP server" guide for the exact setup.

In practice, your assistant stops guessing about Firebase or Google Cloud APIs and starts looking them up.

In your own code

If you are building something that answers questions about Google products, like an internal help bot or a CI step, the client libraries are the route. They support C#, Go, Java, Node.js and TypeScript, PHP, Python and Ruby.

Google lists what you get on top of raw HTTP: native types, builder patterns, error handling and async execution per language, plus Application Default Credentials, automatic retries and structured citation tracking.

Here is Google's own Python example, which fetches the latest release notes for a product so an agent can answer past its training cutoff:

from google.cloud import developer_knowledge_v1

def FetchMyProductReleaseNotes(my_product_name: str) -> str:
    """Fetches the latest release notes to bridge a model's knowledge cutoff."""
    # Initialize the Developer Knowledge API client
    my_api_client = developer_knowledge_v1.DeveloperKnowledgeClient()

    # Formulate a targeted query for the specific product's release notes
    my_search_query = f"{my_product_name} latest release notes"
    my_request = developer_knowledge_v1.SearchDocumentChunksRequest(
        query=my_search_query,
    )

    # Retrieve and return the relevant documentation chunks
    my_response = my_api_client.search_document_chunks(request=my_request)

    # Extract the chunk text
    my_release_notes = [my_chunk.text for my_chunk in my_response.chunks]
    return "\n".join(my_release_notes)

# Example usage:
# my_latest_notes = FetchMyProductReleaseNotes("Firestore")

About twenty lines, and your agent has a tool that reads current release notes instead of remembering old ones.

In the browser

If you want to see what the responses look like before writing any code, the Google APIs Explorer page for the Developer Knowledge API lets you send requests and inspect the response structure with zero code.

What does the announcement not tell you?

It says the API covers "Google Cloud, Firebase, Android, and more," but does not list every product. It does not mention pricing, quotas or rate limits. It gives no number for what "indexed frequently" means.

If any of those matter to your use case, read the quickstart and docs before you put a production pipeline on top of it. I would treat it as something to try now and depend on once you have checked those details.

What should you do this week?

Pick the path that matches how you already work. Google's own next steps split it the same way: if you have the gcloud CLI, start querying; if you don't, give the agent skill to your assistant.

1. If you use gcloud: run gcloud developer-knowledge answer-query on one question you looked up by hand this week. Compare the answer to what you found.
2. If you use Claude Code, Cursor or Copilot: run the npx skills add command in one project that touches Firebase or Google Cloud. Ask your assistant a question it got wrong before.
3. Next time you hit an error: save the trace to error.txt and pipe it into answer-query. That is the fastest test of whether this is useful for you.
4. If you build internal tools: open the APIs Explorer and look at one SearchDocumentChunks response. Check whether the chunks and citations fit what your bot needs.
5. Watch the token cost: for one task, compare how much context your assistant uses with the skill versus pasting a docs page in yourself.

The first three take an afternoon at most. That is enough to know whether this earns a place in your setup.

FAQ

What is the Google Developer Knowledge API?

It is Google's official, programmatic source for developer documentation about Google Cloud, Firebase, Android and more. It returns fresh, Markdown-formatted docs and supports semantic and keyword search, document chunking and grounded Q&A. Google built it for AI agents, IDE extensions and automated workflows.

Does the Developer Knowledge MCP server work with Claude Code and Cursor?

Yes, according to Google. The agent skill uses standard MCP integration, and Google names Google Antigravity, Claude Code, Cursor, GitHub Copilot and custom agentic frameworks. You install it with npx skills add google/skills --skill retrieving-developer-knowledge, and it can fall back to the REST API via curl.

Which languages have client libraries?

Seven: C#, Go, Java, Node.js and TypeScript, PHP, Python and Ruby. They expose AnswerQuery, SearchDocumentChunks, GetDocument and BatchGetDocuments. The batch method retrieves up to 20 documents in one call.

Do I need to install anything to use it from the terminal?

Not if you already have the Google Cloud SDK. Google says the gcloud surface comes pre-installed in Cloud Shell and works on standard SDK installs on Linux, macOS and Windows without extra plugins or configuration.

Install the skill in one project today and ask it the question your assistant last got wrong. Then browse the free tools on Art of Code, and if you are watching what all that context costs you, read our breakdown of why Claude cache reads just got 75% cheaper.

Written by Adithya Guttha, Founder of Art of Code.