Web

Search and fetch content from the web. The provider exposes one tool: query_web. You choose the backend (Exa, Parallel, or MCP).

from agno.agent import Agent
from agno.context.web import WebContextProvider, ExaBackend

web = WebContextProvider(backend=ExaBackend())

agent = Agent(
    model=...,
    tools=web.get_tools(),
)

await agent.aprint_response("What are the latest developments in AI agents?")

WebContextProvider is read-only. There is no update_web tool.

Installation

Install the optional dependency for the selected backend.

Backend	Install command
`ExaBackend`	`uv pip install "agno[exa]"`
`ParallelBackend`	`uv pip install "agno[parallel]"`
`ExaMCPBackend` or `ParallelMCPBackend`	`uv pip install "agno[mcp]"`

Backends

Exa
Exa MCP
Parallel
Parallel MCP

Neural search engine. Best for semantic queries.

from agno.context.web import WebContextProvider, ExaBackend

web = WebContextProvider(backend=ExaBackend())

Requires EXA_API_KEY environment variable.

Exa via MCP protocol.

from agno.context.web import WebContextProvider, ExaMCPBackend

web = WebContextProvider(backend=ExaMCPBackend())

Requires lifecycle setup with await web.asetup().

Web search and extraction via Parallel’s API.

from agno.context.web import WebContextProvider, ParallelBackend

web = WebContextProvider(backend=ParallelBackend())

Requires PARALLEL_API_KEY environment variable.

Parallel via MCP protocol.

from agno.context.web import WebContextProvider, ParallelMCPBackend

web = WebContextProvider(backend=ParallelMCPBackend())

Requires lifecycle setup with await web.asetup().

Configuration

Parameter	Type	Default	Description
`backend`	`ContextBackend`	required	Search backend (ExaBackend, ExaMCPBackend, ParallelBackend, ParallelMCPBackend).
`id`	`str`	`"web"`	Tool becomes `query_<id>`.
`model`	`Model`	`None`	Model for the sub-agent.
`mode`	`ContextMode`	`default`	See Mode.

Tools Exposed

Tool	Description
`query_web`	Search the web, fetch pages, synthesize answers with citations.

Lifecycle

MCP backends (ExaMCPBackend, ParallelMCPBackend) require setup and teardown for the server connection. The SDK backends need neither:

web = WebContextProvider(backend=ExaMCPBackend())

await web.asetup()
try:
    agent = Agent(model=..., tools=web.get_tools())
    await agent.arun("Search for...")
finally:
    await web.aclose()

Example queries

Query	What happens
”What is the current state of WebGPU support?”	Searches, fetches recent articles, synthesizes
”Find documentation on Python 3.12 new features”	Searches docs, returns summary with links
”Research competitors to Stripe Atlas”	Multi-source search and synthesis

Get Started

Basics

Advanced

Production

Providers

Other

Additional Resources

Installation

Backends

Configuration

Tools Exposed

Lifecycle

Example queries

Resources

Exa Tools Reference

Cookbook Example

​Installation

​Backends

​Configuration

​Tools Exposed

​Lifecycle

​Example queries

​Resources

Exa Tools Reference

Cookbook Example

Installation

Backends

Configuration

Tools Exposed

Lifecycle

Example queries

Resources