get_full_content_tool

Retrieve complete documentation page content by URL and version

Overview

The get_full_content_tool retrieves the complete, reassembled content of a documentation page. Use this tool when search results provide a relevant chunk, but you need to see the full page context.

This tool is typically called after search_documents_tool returns results. Each search result includes a tool_hint with the exact parameters to pass to this tool.

Parameters

url · string · required

The complete URL of the documentation page to retrieve. Must match exactly with a URL in the database.

Example: “https://fastapi.tiangolo.com/tutorial/security/first-steps/”

version · string · required

The version of the documentation. Must match the version string stored in the database.

Example: “0.104.0”, “latest”, “v4.5.2”

Return Format

Returns a formatted markdown string containing the complete page content.

Successful Retrieval

# {page_title}

Source: {url}
Version: {version}

{full_content_chunk_1}

{full_content_chunk_2}

{full_content_chunk_3}
...

Page Not Found

No content found for URL: {url} (version: {version})

Database Error

No content found for URL: {url}

Response Fields

The formatted response includes:

title · string

The title of the documentation page (from the first chunk’s metadata)

source · string

The original URL of the documentation page

version · string

The version identifier for this documentation snapshot

content · string

The complete page content, reassembled from all chunks in correct order (sorted by chunk_index)

How It Works

Chunk Reassembly

Documentation pages are stored as chunks during ingestion:

1. Query database: Find all chunks matching the URL and version
2. Sort by chunk_index: Ensures content is in original order
3. Concatenate: Joins chunks with double newlines (\n\n)
4. Format: Wraps in markdown with title and metadata

Database Query

SELECT title, content, chunk_index
FROM documents
WHERE url = '{url}' AND version = '{version}'
ORDER BY chunk_index ASC

Chunking Context

During ingestion, pages are split into overlapping chunks:

• Default chunk size: 800 characters
• Default overlap: 200 characters
• Chunk index: Sequential integer starting at 0

When reassembled, the overlap is preserved, which may result in some repeated content at chunk boundaries.

Example Usage

From Search Result

Typical workflow:

// Step 1: Search
{
  "query": "authentication middleware",
  "library_name": "fastapi",
  "version": "0.104.0"
}

// Response includes:
// To get full page content: {"tool": "get_full_content", "url": "https://fastapi.tiangolo.com/tutorial/security/first-steps/", "version": "0.104.0"}

// Step 2: Get full content
{
  "url": "https://fastapi.tiangolo.com/tutorial/security/first-steps/",
  "version": "0.104.0"
}

Response:

# Security - First Steps

Source: https://fastapi.tiangolo.com/tutorial/security/first-steps/
Version: 0.104.0

Let's imagine that you have your backend API in some domain.

And you have a frontend in another domain or in a different path of the same domain (or in a mobile application).

And you want to have a way for the frontend to authenticate with the backend, using a username and password.

We can use OAuth2 to build that with FastAPI.

...

[complete page content continues]

Direct URL Access

{
  "url": "https://react.dev/reference/react/useState",
  "version": "18.2.0"
}

Response:

# useState

Source: https://react.dev/reference/react/useState
Version: 18.2.0

useState is a React Hook that lets you add a state variable to your component.

const [state, setState] = useState(initialState)

...

[complete page content]

Page Not Found

{
  "url": "https://fastapi.tiangolo.com/nonexistent-page/",
  "version": "0.104.0"
}

Response:

No content found for URL: https://fastapi.tiangolo.com/nonexistent-page/ (version: 0.104.0)

Integration Patterns

Parse Tool Hint from Search

import json

# After search_documents_tool returns results
search_response = """
Found 2 matches.
1. **Security**: "OAuth2 with Password..." (Source: https://..., Version: 0.104.0)
   To get full page content: {"tool": "get_full_content", "url": "https://fastapi.tiangolo.com/tutorial/security/first-steps/", "version": "0.104.0"}
"""

# Extract and parse the tool hint
import re
tool_hint_match = re.search(r'To get full page content: ({.*?})', search_response)
if tool_hint_match:
    hint = json.loads(tool_hint_match.group(1))
    
    # Call get_full_content_tool with parsed parameters
    full_content = get_full_content_tool(
        url=hint["url"],
        version=hint["version"]
    )

Selective Full Content Retrieval

# Only fetch full content for the most relevant result
if search_results_count > 0:
    # Get tool hint from first result
    first_result_hint = extract_first_tool_hint(search_response)
    
    full_content = get_full_content_tool(
        url=first_result_hint["url"],
        version=first_result_hint["version"]
    )

Batch Content Retrieval

# Get full content for multiple relevant results
search_results = parse_search_results(search_response)

for result in search_results[:3]:  # Top 3 results
    full_content = get_full_content_tool(
        url=result["url"],
        version=result["version"]
    )
    
    # Process each full page
    analyze_content(full_content)

Best Practices

Use tool hints from search results:

Don’t manually construct URLs. Always use the tool hint provided in search_documents_tool results to ensure exact URL matching.

URL must match exactly:

URLs are matched byte-for-byte. Ensure no extra/missing slashes, query parameters, or fragments unless they were in the original indexed URL.

When to Fetch Full Content

Fetch full content when:

• Search chunk doesn’t provide enough context
• You need to see complete code examples
• Multiple sections of the page might be relevant
• User explicitly asks to “see the full page”

Don’t fetch full content when:

• Search chunk already answers the question
• You’re just verifying a quick fact
• Multiple results need triage first
• Bandwidth/latency is a concern

Content Size Awareness

Full pages can be large:

Page Type	Typical Size	Chunk Count
API reference	5-15KB	8-20 chunks
Tutorial	10-30KB	15-40 chunks
Guide	15-50KB	20-70 chunks
Long reference	50-200KB	70-300 chunks

Consider token limits when passing full content to LLMs.

Performance Characteristics

Query Performance

• Small pages (< 5 chunks): < 10ms
• Medium pages (5-20 chunks): 10-50ms
• Large pages (20-100 chunks): 50-200ms
• Very large pages (> 100 chunks): 200-500ms

Caching

Unlike list_libraries_tool, this tool does not cache results. Each call performs a fresh database query.

Rationale:

• Full content is typically fetched once per page
• Caching would consume significant memory
• Database queries are fast enough for on-demand retrieval

Troubleshooting

”No content found” Errors

Cause 1: URL mismatch

Stored:  https://example.com/docs/page/
Queried: https://example.com/docs/page
         (missing trailing slash)

Solution: Use exact URL from search result tool hint

---

Cause 2: Version mismatch

Stored:  version="latest"
Queried: version="1.0.0"

Solution: Use exact version from search result tool hint

---

Cause 3: Page never indexed

The page might not have been included during library ingestion.

Solution: Re-run ingestion with updated sitemap/crawler settings

Truncated or Incomplete Content

Symptom: Content seems to cut off mid-sentence

Cause: Chunks were not properly indexed or chunk_index is missing

Solution:

1. Check ingestion logs for errors
2. Re-ingest the library version
3. Verify chunk_index values in database

Duplicate Content at Chunk Boundaries

Symptom: Some sentences/paragraphs appear twice

Cause: Chunk overlap during ingestion (this is expected behavior)

Solution: This is normal. The overlap ensures context is preserved across chunk boundaries. If it’s excessive, reduce chunk_overlap configuration before re-ingesting.

Wrong Content Returned

Symptom: Content doesn’t match the URL

Cause: Database corruption or multiple versions with same URL

Solution:

1. Verify version parameter is correct
2. Check for duplicate entries: openground list <library>
3. Delete and re-ingest the library version