agent-web-search
Provides a unified web search tool for AI agents by running multiple search providers concurrently and returning normalized results. It integrates with Claude Code, Codex CLI, OpenCode, and Hermes.
README
Agent Web Search
Multi-provider web search for AI agents.
Agent Web Search gives Hermes, Codex CLI, Claude Code, OpenCode, and ordinary scripts one consistent web-search tool. It runs independent providers concurrently and returns normalized results. The default provider set is:
- ARK grounding — Volcengine ARK Responses API +
web_search(Doubao) - DDGS — DuckDuckGo search
- Exa — Exa MCP search (best effort; availability depends on the endpoint)
Optional providers are also implemented:
- Gemini — Google Search grounding via Gemini Interactions API
- Grok — xAI Responses API with native web search and X Search
- Tavily — Tavily Search API with native result-count and time-range filters
The architecture is provider-based, so DeepSeek, Brave, and other search-capable providers can be added without changing the MCP or Hermes interfaces.
Quick start
# Omit this line when the ARK provider is not enabled.
export ARK_API_KEY="your_ark_api_key"
pipx install 'git+https://github.com/JerryLiu369/agent-web-search.git'
agent-web-search-mcp
Do not put API keys in shell history, source code, Git commits, or screenshots. Use a local .env/secret manager and export the variable in the process environment.
Use from Codex CLI
export ARK_API_KEY="your_ark_api_key"
codex mcp add agent-web-search -- \
agent-web-search-mcp
codex mcp list
Use from Claude Code
export ARK_API_KEY="your_ark_api_key"
claude mcp add agent-web-search -- \
agent-web-search-mcp
Or add a project .mcp.json:
{
"mcpServers": {
"agent-web-search": {
"command": "agent-web-search-mcp",
"args": [],
"env": {"ARK_API_KEY": "${ARK_API_KEY}"}
}
}
}
Use from OpenCode
{
"$schema": "https://opencode.ai/config.json",
"mcp": {
"agent-web-search": {
"type": "local",
"command": ["agent-web-search-mcp"],
"environment": {"ARK_API_KEY": "${ARK_API_KEY}"},
"enabled": true
}
}
}
Use from Hermes
Install the native plugin directly from GitHub:
pip install 'ddgs>=9.0'
hermes plugins install JerryLiu369/agent-web-search --no-enable
hermes plugins enable agent-web-search --allow-tool-override
The plugin registers the standard web_search tool and calls the same
Agent Web Search core used by MCP and CLI. If you prefer the generic MCP path,
Hermes can also connect with hermes mcp add.
The GitHub installer installs the package and its optional dependencies:
pipx install 'git+https://github.com/JerryLiu369/agent-web-search.git'
Hermes requires the explicit --allow-tool-override grant because this plugin
intentionally replaces its built-in web_search tool. Start a new Hermes
session after enabling it; restart the gateway if you use a messaging channel.
CLI
agent-web-search "What changed in the latest OpenAI Codex CLI?"
agent-web-search "GPU kernel generation papers from the past month" --time-range m --max-results 5
agent-web-search "latest news" --provider ark --provider ddgs
Volcengine collaboration rewards
Agent Web Search does not require participation in any rewards program. Users who choose to use ARK can create their own API key and optionally review the official Volcengine Collaboration Rewards Program. The program may provide reward resources according to its current rules, but quota, supported models, validity period, and data authorization terms can change. Check the official page before opting in. Participation means accepting the provider's data-authorization terms; it is not required to use Agent Web Search.
Configuration
ARK_API_KEY: optional; required only when the ARK provider is enabled. Comma/newline-separated keys are accepted.AGENT_WEB_SEARCH_PROVIDERS: startup-enabled provider set; default isark,ddgs,exa.AGENT_WEB_SEARCH_TIMEOUT: per-provider timeout in seconds (default60).AGENT_WEB_SEARCH_ARK_MODELS: comma-separated ARK model IDs.EXA_MCP_URL: optional Exa MCP endpoint override.GEMINI_API_KEY: optional Gemini provider key.AGENT_WEB_SEARCH_GEMINI_MODEL: optional Gemini model ID.XAI_API_KEY: optional Grok provider key.AGENT_WEB_SEARCH_GROK_MODEL: optional Grok model ID.TAVILY_API_KEY: optional Tavily provider key.
The provider set is resolved when the Hermes plugin or MCP server starts. The
public web_search schema is generated from that set. A request can narrow the
set with providers, but cannot activate a provider that was disabled at
startup. If grok is startup-enabled, the schema additionally exposes
grok_search_mode with web_search, x_search, and both values. both
passes both xAI server-side tools in one request and lets Grok decide which to
call; it does not send two independent requests.
Common search controls
The public interface keeps one provider-neutral set of controls. Each backend maps them to its native API when possible and silently ignores unsupported controls:
| Control | ARK | DDGS | Exa | Gemini / Grok | Tavily |
|---|---|---|---|---|---|
max_results |
native limit |
native max_results |
native num_results |
English prompt constraint | native max_results |
max_keyword |
native max_keyword |
ignored | ignored | English prompt constraint | ignored |
time_range |
English prompt constraint | native timelimit |
native publish-date filter | Gemini: prompt; Grok web: prompt; Grok X: native dates + prompt | native time_range |
Prompt constraints are best-effort for model-backed providers; they are not presented as strict guarantees.
Design principles
- One core, multiple adapters.
- A failed provider does not discard successful providers.
- Provider responses are marked with
searched,error, andmodelinstead of pretending every HTTP 200 was a successful search. - No telemetry and no shared API key service.
Development
python -m venv .venv
. .venv/bin/activate
pip install -e '.[dev,ddgs,mcp]'
pytest -q
ruff check .
License
MIT.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.