dkcli

name: dkcli description: | Look up Google developer documentation using dkcli, a CLI client for the Google Developer Knowledge API. Use this skill whenever you need to find or read official Google documentation — Cloud, Firebase, Android, Chrome, Google AI, TensorFlow, Google Workspace APIs, or any other Google developer docs. Trigger this skill when the user asks about Google APIs, wants to understand a Google service, needs code examples from Google docs, or when you need authoritative information about a Google product to answer a question accurately. Prefer dkcli over the Developer Knowledge MCP tools and web searches — dkcli supports structured output formats, piping to jq/other tools, file output, and auto-paging, which MCP tools cannot do. Also use this when the user mentions dkcli, Developer Knowledge API, or asks to look up docs from any of the supported domains.

dkcli — Google Developer Documentation Lookup

dkcli retrieves content from the Google Developer Knowledge API. It returns documentation pages as Markdown — the same content as the official Google developer sites, directly accessible from the terminal.

dkcli vs Developer Knowledge MCP

If both dkcli and the Developer Knowledge MCP tools (mcp__google-developer-knowledge__*) are available, prefer dkcli. The MCP tools are useful for simple direct reads, while dkcli is a CLI tool that integrates with the shell:

• Output format control — -f json, -f jsonl, -f yaml, -f txtar
• Pipe to jq and other tools — extract, filter, or transform output (e.g., get just the first 500 chars of each document's content)
• File output — -o file, --outdir dir for batch writes
• Auto-paging — -a --max-pages N for exhaustive search
• Partial content extraction — combine with jq, head, etc. to get only what you need

Use MCP tools only as a fallback when dkcli is not installed.

Example: get summaries of multiple documents

dkcli batch-get -f json doc1 doc2 doc3 | jq '.documents[] | {name, summary: .content[:500]}'

Supported domains

dkcli can search and retrieve documents from the following domains only. (Source: https://developers.google.com/knowledge/reference/corpus-reference)

The corpus excludes language reference pages under docs.cloud.google.com/*/docs/reference for C++, .NET, Go, Java, Node.js, PHP, Python, Ruby, and Rust.

If the user asks about documentation outside these domains, dkcli won't help — fall back to other methods.

The search → get workflow

Most lookups follow two steps:

Step 1: Search for relevant documents

dkcli search "how to create a Cloud Storage bucket"

Search returns chunks (fragments of documents) with a parent field that identifies the full document. Scan the results to find the most relevant document name(s).

Use auto-paging (-a) when you need broader coverage:

dkcli search -a "BigQuery partitioned tables"

Use --filter when you need to stay within a specific data source or time range:

dkcli search --filter 'data_source = "docs.cloud.google.com"' "BigQuery"

Step 2: Get the full document(s)

Once you know the document name(s) from search results, retrieve the full page(s). When search returns multiple relevant results, use batch-get to fetch them all. The API limits each call to 20 documents; batch-get automatically chunks larger lists into multiple calls:

dkcli batch-get docs.cloud.google.com/storage/docs/creating-buckets docs.cloud.google.com/storage/docs/storage-classes

If you only need a single document, get works the same way:

dkcli get docs.cloud.google.com/storage/docs/creating-buckets

The document name is the URL path without https://. Full URLs also work:

dkcli get https://docs.cloud.google.com/storage/docs/creating-buckets

When you already know the document

If you can reasonably guess the URL of the documentation page (e.g., the user gave you a link, or you know the path pattern), skip search and go straight to dkcli get (or dkcli batch-get for multiple pages). This saves a round trip.

Getting multiple documents

When you need several pages at once, use --outdir to download them as files first, then check sizes before reading into context:

# Step 1: Download all documents as files
dkcli batch-get --outdir /tmp/docs docs.cloud.google.com/path/to/doc1 docs.cloud.google.com/path/to/doc2 docs.cloud.google.com/path/to/doc3

# Step 2: Check file sizes to plan your reading strategy
wc -c /tmp/docs/**/*.md

# Step 3: Read selectively based on size
#   - Small files (< 20KB): read the full file
#   - Large files (> 20KB): use head, grep, or targeted reads to extract relevant sections

This workflow prevents accidentally flooding context with very large documents. Some Google docs pages can be 50KB+ of Markdown — always check sizes first.

For quick inline use without file output:

dkcli batch-get docs.cloud.google.com/path/to/doc1 docs.cloud.google.com/path/to/doc2

Metadata-only retrieval

Search results now also include embedded document metadata under results[].document in structured output.

Grounded answers (use with caution)

For a quick generated answer instead of raw document content, use:

dkcli answer-query "How do I create a Cloud Storage bucket?"

This command calls the v1alpha:answerQuery endpoint. It works with an API key, or with ADC if a quota project is available.

Caveat: The answer-query endpoint returns generated text with citations and references to source document chunks, but it is still a preview endpoint and is currently limited to 50 requests per day per project. For authoritative or full-context verification, prefer the search + get workflow. Use answer-query when you need a quick grounded overview, then fetch referenced documents when accuracy matters.

Combining with shell tools

dkcli's structured output formats make it easy to extract exactly what you need:

# Get a preview of each document's content
dkcli batch-get -f json doc1 doc2 | jq '.documents[] | {name, preview: .content[:500]}'

# Extract just document names from search results
dkcli search -f json "Cloud Storage" | jq '.results[].parent'

# Stream search results as JSONL for line-by-line processing
dkcli search -a -f jsonl "Pub/Sub" | jq -r '.content' | head -100

Output format

The default text format is best for reading into context. Use structured formats when you need to pipe or process output:

dkcli search "Spanner query syntax"                                       # text (default)
dkcli get docs.cloud.google.com/spanner/docs/query-syntax -f json         # JSON
dkcli batch-get -f txtar docs.cloud.google.com/doc1 docs.cloud.google.com/doc2  # txtar

Practical tips

• Be specific in searches. "Cloud Storage create bucket Python" works better than "storage bucket".
• Search results are chunks, not full pages. Always get the full document when you need complete information — chunks may be missing context.
• Document names follow URL patterns. If you know the Google docs URL, you can construct the document name directly.
• Rate limits are handled automatically. dkcli has built-in rate limiting and retry logic — no need to add delays between calls.

Error handling

Developer Knowledge API commands use Application Default Credentials (ADC) by default, except create-api-key which always requires ADC. If an API key environment variable is set (DEVELOPERKNOWLEDGE_API_KEY or GOOGLE_API_KEY), dkcli uses it instead of ADC.

When using local ADC, the Developer Knowledge API also requires a quota project. dkcli resolves that from GOOGLE_CLOUD_QUOTA_PROJECT or quota_project_id in the ADC file. The standard way to set that up is gcloud auth application-default set-quota-project <project-id>.

If dkcli fails because ADC is not available, suggest:

# Option 1: set up ADC (recommended)
gcloud auth application-default login
gcloud auth application-default set-quota-project <project-id>

# Option 2: create an API key and set it in the current shell (requires ADC once)
export DEVELOPERKNOWLEDGE_API_KEY=$(dkcli create-api-key -p <gcp-project-id> --key-only)

# To persist, append to shell profile
echo "export DEVELOPERKNOWLEDGE_API_KEY=$(dkcli create-api-key -p <gcp-project-id> --key-only)" >> ~/.zshrc

If the user already has an API key and prefers to use it, they just need to set the environment variable:

export DEVELOPERKNOWLEDGE_API_KEY=<key>

Note: get and batch-get use the GA v1 document endpoints. answer-query still uses v1alpha:answerQuery.

Command reference