Docs: add an AG2 integration page (AG2 ships ExaToolkit built on exa-py) #431

Why this is filed here

The same request is open as exa-py#252 and has had no maintainer response in three weeks. Filing it in this repo, which is the most-starred repo in the org, in the hope it reaches whoever owns the docs site. Happy to have it closed as a duplicate of #252 — the ask is identical, and it is docs-only: nothing needs to be built or maintained on the Exa side.

Summary

AG2 now ships a first-class Exa integration built on exa-py, but there's no AG2 page in the Exa docs (there are ones for LangChain, CrewAI, LlamaIndex, Google ADK, …). We'd like to get one added, and we're happy to provide the content.

What landed on the AG2 side

As of ag2>=1.0.0 (current release 1.0.4), AG2 ships ExaToolkit in ag2.extensions.tools.search. It wraps exa_py.AsyncExa and exposes four Exa endpoints as agent tools sharing a single client:

Each tool is also a factory method (toolkit.search(), toolkit.answer(), …), so per-tool Exa filters — search type, category, include/exclude domains, published/crawl date ranges, livecrawl, num_results, max_characters — can be bound per agent, and an agent can be given only the tools it needs.

pip install "ag2>=1.0.0" "exa-py>=2.12.1,<3"

import os

from ag2 import Agent
from ag2.config import AnthropicConfig
from ag2.extensions.tools.search import ExaToolkit

agent = Agent(
    "researcher",
    config=AnthropicConfig(model="claude-sonnet-4-6"),
    tools=[ExaToolkit(api_key=os.environ["EXA_API_KEY"])],
)

reply = await agent.ask("Find recent papers on neural retrieval and summarise them.")

Links:

• AG2 docs page: https://docs.ag2.ai/docs/user-guide/extensions/tools/search/exa/
• Source: ag2/extensions/tools/search/exa.py in https://github.com/ag2ai/ag2

The ask

Add an AG2 page to the Exa docs — exa.ai/docs/integrations/ag2 (or /docs/reference/ag2, whichever matches your current grouping) plus a nav entry alongside the other agent-framework integrations.

We can't open the PR ourselves: exa.ai/docs is deployed from the private exa-labs/monorepo repo (docs/ directory), so this issue is the closest thing to one. Below is a ready-to-drop .mdx written in the style of your existing integration pages — feel free to edit or trim it.

---
title: "AG2"
description: "Use Exa search, find-similar, contents, and answer as tools in AG2 agents with the built-in ExaToolkit."
---

[AG2](https://github.com/ag2ai/ag2) is an open-source framework for building agents and multi-agent
applications. Since `v1.0.0`, AG2 ships `ExaToolkit` out of the box: it is built directly on the
[Exa Python SDK](https://github.com/exa-labs/exa-py) and exposes four Exa endpoints as agent tools
that share a single `AsyncExa` client.

## Prerequisites

* Create an [API key](https://dashboard.exa.ai/api-keys) in Exa and set it as `EXA_API_KEY`.

```bash theme={null}
pip install "ag2>=1.0.0" "exa-py>=2.12.1,<3"

import os

from ag2 import Agent
from ag2.config import AnthropicConfig
from ag2.extensions.tools.search import ExaToolkit

agent = Agent(
    "researcher",
    config=AnthropicConfig(model="claude-sonnet-4-6"),
    tools=[ExaToolkit(api_key=os.environ["EXA_API_KEY"])],
)

reply = await agent.ask("Find recent papers on neural retrieval and summarise them.")

toolkit = ExaToolkit(api_key=os.environ["EXA_API_KEY"])

agent = Agent(
    "researcher",
    config=config,
    tools=[toolkit.search(), toolkit.answer()],
)

search_tool = toolkit.search(
    num_results=5,
    max_characters=2000,        # returns full text via search_and_contents
    search_type="neural",       # "auto" | "neural" | "fast" | "deep-lite" | "deep" | ...
    category="research paper",
    include_domains=["arxiv.org"],
    start_published_date="2024-01-01",
    livecrawl="always",
)

</details>

### Also happy to do

If it's useful, we can open a small PR against this repo's `README.md` too: the "Other MCP Clients" table lists editors and desktop clients only, and AG2 can point at `https://mcp.exa.ai/mcp` directly through its built-in `MCPToolkit` — a one-row addition. Just say the word and we'll send it.