Content Core

Extract, process, and summarize content from URLs, files, and text through a unified async Python API, CLI, or MCP server.

Supported Formats

Quick Start

pip install content-core

import content_core

result = await content_core.extract_content(url="https://example.com")
print(result.content)

Or with zero install:

uvx content-core extract "https://example.com"

CLI Usage

Content Core provides a unified content-core command with subcommands for extraction, summarization, and MCP server.

Extract

# From a URL
content-core extract "https://example.com"

# From a file
content-core extract document.pdf

# With JSON output
content-core extract document.pdf --format json

# With a specific engine
content-core extract "https://example.com" --engine firecrawl

# From stdin
echo "some text" | content-core extract

Summarize

# Summarize text
content-core summarize "Long article text here..."

# With context
content-core summarize "Long text" --context "bullet points"

# From stdin
cat article.txt | content-core summarize --context "explain to a child"

MCP Server

content-core mcp

Configuration

# Set persistent config
content-core config set llm_provider anthropic
content-core config set llm_model claude-sonnet-5

# List current config
content-core config list

# Delete a config value
content-core config delete llm_provider

Config is stored in ~/.content-core/config.toml. Priority: command flags > env vars > config file > defaults.

Zero-Install with uvx

All commands work without installation using uvx:

uvx content-core extract "https://example.com"
uvx content-core summarize "text" --context "one sentence"
uvx content-core mcp

Python API

Extraction

import content_core

# From a URL
result = await content_core.extract_content(url="https://example.com")

# From a file
result = await content_core.extract_content(file_path="document.pdf")

# From text
result = await content_core.extract_content(content="some text")

# With engine override
from content_core import ContentCoreConfig
config = ContentCoreConfig(url_engine="firecrawl")
result = await content_core.extract_content(url="https://example.com", config=config)

Summarization

import content_core

summary = await content_core.summarize("long article text", context="bullet points")

Configuration

from content_core import ContentCoreConfig

config = ContentCoreConfig(
    url_engine="firecrawl",
    document_engine="docling",
    audio_concurrency=5,
)
result = await content_core.extract_content(url="https://example.com", config=config)

MCP Integration

Content Core includes a Model Context Protocol (MCP) server for use with Claude Desktop and other MCP-compatible applications.

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "content-core": {
      "command": "uvx",
      "args": ["content-core", "mcp"],
      "env": {
        "OPENAI_API_KEY": "sk-..."
      }
    }
  }
}

The MCP server exposes two tools: extract_content and summarize_content. Both return plain text.

For detailed setup, see the MCP documentation.

Agent Skill (Claude Code & Codex)

Content Core ships an Agent Skill that teaches AI agents how to use it for extracting content from external sources. This repository is also a plugin marketplace, so the skill installs natively in both harnesses.

Claude Code — add the marketplace and install the plugin:

/plugin marketplace add lfnovo/content-core
/plugin install content-core@content-core

Codex — the repository carries a Codex plugin manifest (.codex-plugin/plugin.json) and marketplace catalog (.agents/plugins/marketplace.json) pointing at the same skill.

Manual fallback — copy the skill file directly into your project:

curl -o .claude/skills/content-core/SKILL.md --create-dirs \
  https://raw.githubusercontent.com/lfnovo/content-core/main/skills/content-core/SKILL.md

Once installed, the agent can use content-core to extract content from URLs, documents, and media files — either via CLI (uvx content-core) or MCP if configured.

AI Providers

Content Core uses Esperanto to support multiple LLM and STT providers. Switch providers by changing the config — no code changes needed:

# Use Anthropic for summarization
content-core config set llm_provider anthropic
content-core config set llm_model claude-sonnet-5

# Use Groq for transcription
content-core config set stt_provider groq
content-core config set stt_model whisper-large-v3

Supported providers include OpenAI, Anthropic, Google, Groq, DeepSeek, Ollama, and more. See the Esperanto documentation for the full list.

Configuration

Content Core uses ContentCoreConfig powered by pydantic-settings. Settings are resolved in priority order: constructor args > env vars (CCORE_*) > config file (~/.content-core/config.toml) > defaults.

Environment Variables

API keys for external services are set via their standard environment variables (e.g., OPENAI_API_KEY, FIRECRAWL_API_KEY, JINA_API_KEY).

Proxy Configuration

Content Core reads standard HTTP_PROXY / HTTPS_PROXY / NO_PROXY environment variables automatically. No additional configuration is needed.

YouTube on Blocked Networks

If YouTube extraction fails with IpBlocked/RequestBlocked on a flagged IP (cloud hosts, VPNs), either point CCORE_YOUTUBE_COOKIES_FILE at a cookies.txt exported from a logged-in browser (re-export it when the cookies expire), or set CCORE_YOUTUBE_PROXY to a residential proxy — datacenter proxies get blocked or CAPTCHA'd. See docs/usage.md.

Optional Dependencies

# Docling for advanced document parsing (PDF, DOCX, PPTX, XLSX)
# Required by document_engine="docling", which raises ConfigurationError without it.
# Use document_engine="auto" (default) or "simple" to proceed without Docling.
pip install content-core[docling]

# Crawl4AI for local browser-based URL extraction
pip install content-core[crawl4ai]
python -m playwright install --with-deps

# LangChain tool wrappers
pip install content-core[langchain]

# All optional features
pip install content-core[docling,crawl4ai,langchain]

Using with LangChain

When installed with the langchain extra, Content Core provides LangChain-compatible tool wrappers:

from content_core.tools import extract_content_tool, summarize_content_tool

tools = [extract_content_tool, summarize_content_tool]

Documentation

• Usage Guide -- Python API details, configuration, and examples
• Processors -- How content extraction works for each format
• MCP Server -- Claude Desktop and MCP integration

Development

git clone https://github.com/lfnovo/content-core
cd content-core

uv sync --group dev

# Run tests
make test

# Lint
make ruff

License

This project is licensed under the MIT License.

Contributing

Contributions are welcome! Please see our Contributing Guide for details.