repository·main·Indexed 23 days ago
https://github.com/crewaiinc/crewai-toolsA library of tools designed to extend crewAI agents with capabilities such as file management, web scraping, database access, and AI-powered tools. It includes native support for the Model Context Protocol (MCP) and specialized toolkits for AWS Bedrock (Browser, Code Interpreter, and Knowledge Base Retriever) and AWS S3 (S3ReaderTool and S3WriterTool).
The BrightData tool suite provides three main capabilities for CrewAI agents:
The FileCompressorTool allows you to compress individual files or entire directories (including nested subdirectories) into various archive formats. It supports .zip, .tar, .tar.gz, .tar.bz2, and .tar.xz formats.
Key features include:
overwrite flag.from crewai_tools import FileCompressorTool
tool = FileCompressorTool()
# Example: Compress a directory into a zip archive
result = tool._run(
input_path="./data/project_docs",
output_path="./output/project_docs.zip",
overwrite=True
)
print(result)EmbedChain to allow CrewAI agents to answer questions by querying information retrieved from various data sources. It is highly versatile, supporting a wide range of inputs from local files and directories to web pages, YouTube content, and various SaaS platforms like Gmail, Slack, and GitHub.The tool supports several ways to establish a connection to SingleStore:
host, user, password, database, and port individually.host parameter (e.g., user:password@host:port/database).SINGLESTOREDB_URL in your environment and initialize the tool without arguments.ssl_ca, ssl_cert, and ssl_key for secure connections.# Connection URL method
tool = SingleStoreSearchTool(host='user:password@localhost:3306/database_name')
# Environment Variable method (requires export SINGLESTOREDB_URL=...)
tool = SingleStoreSearchTool()
# SSL method
tool = SingleStoreSearchTool(
host='your_host',
user='your_username',
password='your_password',
database='your_database',
ssl_ca='/path/to/ca-cert.pem',
ssl_cert='/path/to/client-cert.pem',
ssl_key='/path/to/client-key.pem'
)The SerperDevTool returns structured data depending on the search_type used. The response includes:
search_type="search"): Search parameters, knowledge graph data, organic search results (with sitelinks), "People Also Ask" questions, and related searches.search_type="news"): Search parameters and news results containing date, source, and image information.The StagehandTool integrates the Stagehand framework into CrewAI, allowing agents to interact with websites using natural language. It operates via three core primitives:
It is highly recommended to use the tool within a context manager to ensure that browser resources are automatically cleaned up when the agent finishes its task.
from crewai import Agent, Task, Crew
from crewai_tools import StagehandTool
from stagehand.schemas import AvailableModel
# Use a context manager for automatic resource cleanup
with StagehandTool(
api_key="your-browserbase-api-key",
project_id="your-browserbase-project-id",
model_api_key="your-llm-api-key",
model_name=AvailableModel.CLAUDE_3_7_SONNET_LATEST,
) as stagehand_tool:
researcher = Agent(
role="Web Researcher",
goal="Find and summarize information",
tools=[stagehand_tool],
...
)
# ... define tasks and crew
crew.kickoff()To prevent unintentional data loss, the tool uses an overwrite boolean flag.
overwrite=True: The tool will replace any existing file at the output_path.overwrite=False (default): If a file already exists at the output_path, the tool will stop and return an error message indicating the file exists.from crewai_tools import FileCompressorTool
tool = FileCompressorTool()
# This will fail if ./backups/my_data_backup.zip already exists
result = tool._run(
input_path="./my_data",
output_path="./backups/my_data_backup.zip",
overwrite=False
)
# Output: Output zip './backups/my_data_backup.zip' already exists and overwrite is set to False.The tool automatically handles common Snowflake errors including DatabaseError, OperationalError, ProgrammingError, network timeouts, and connection issues. These errors are logged and retried based on your max_retries configuration.
It is a best practice to wrap your tool._run calls in try-except blocks to manage failures in your application logic.
import logging
# Configure logging
logging.basicConfig(
level=logging.INFO,
format='%(asctime)s - %(name)s - %(levelname)s - %(message)s'
)
logger = logging.getLogger(__name__)
async def main():
try:
# ... tool initialization ...
results = await tool._run(query="SELECT * FROM table LIMIT 10")
logger.info(f"Query completed successfully. Retrieved {len(results)} rows")
except Exception as e:
logger.error(f"Query failed: {str(e)}")
raiseLlamaIndexTool acts as a general-purpose wrapper. It bridges the gap between LlamaIndex's RAG/agentic pipelines and CrewAI's agent framework. By wrapping LlamaIndex tools or query engines, you can expose complex retrieval logic as standard tools that a CrewAI agent can call to perform searches or data lookups.When using the Model Context Protocol (MCP):
.content[0].text).The DatabricksQueryTool requires authentication credentials. You can authenticate using one of two methods:
DATABRICKS_CONFIG_PROFILE environment variable to your profile name.DATABRICKS_HOST and DATABRICKS_TOKEN environment variables.Example using direct credentials:
export DATABRICKS_HOST="https://your-workspace.cloud.databricks.com"
export DATABRICKS_TOKEN="dapi1234567890abcdef"You can integrate browser automation into your CrewAI agents by using create_browser_toolkit. This function returns both a toolkit object (used for resource management) and a list of browser_tools that can be assigned directly to an Agent.
Important: Always call toolkit.sync_cleanup() when your process is finished to release browser resources.
from crewai import Agent, Task, Crew, LLM
from crewai_tools.aws.bedrock.browser import create_browser_toolkit
# Create the browser toolkit
toolkit, browser_tools = create_browser_toolkit(region="us-west-2")
# Create the Bedrock LLM
llm = LLM(
model="bedrock/us.anthropic.claude-3-7-sonnet-20250219-v1:0",
region_name="us-west-2",
)
# Create a CrewAI agent that uses the browser tools
research_agent = Agent(
role="Web Researcher",
goal="Research and summarize web content",
backstory="You're an expert at finding information online.",
tools=browser_tools,
llm=llm
)
# ... define tasks and crew ...
result = crew.kickoff()
# Clean up browser resources when done
toolkit.sync_cleanup()