X (Twitter)LinkedInGitHubDiscordYouTube

How to integrate Tavily MCP with DeepSeek Harness

Connect DeepSeek Harness to Tavily MCP. Find latest news about electric vehicles, search for recent ai research papers, and more using natural language, with authentication handled for you.

Get started for freeGet a demo
Tavily logoTavily
Api Key

Tavily offers powerful search and data retrieval from documents, databases, and the web. It helps teams locate and filter information instantly, saving hours on research.

5 Tools

Introduction

DeepSeek Harness (dsh) is an open-source agent runtime from DeepSeek AI, and every part of it is a plugin, so you can swap out the model adapter, the tool registry, or the agent loop. It runs on the Cordis framework and comes with a local Web UI, a headless CLI runner, and a Python SDK. You can point it at DeepSeek V4-Pro and V4-Flash, or at Anthropic, OpenAI, and other OpenAI-compatible providers. It is MIT licensed and still in developer preview.

This guide shows how to connect your Tavily account to DeepSeek Harness, and you can do it with either the Composio Connect CLI or the Composio Connect MCP. The dsh-base bundle gives shell execution as a first-class tool, so the CLI is the easiest path for personal use. It is quick to set up and needs no server config, and the MCP path works well too, since the harness comes with a built-in MCP client.

Also integrate Tavily with

ChatGPT WorkAntigravityOpenAI Agents SDKClaude Agent SDKClaude CodeClaude CoworkCodexKimi CodeGrok BuildCursorVS CodeOpenCodeOpenClawHermesPiCLIGoogle ADKLangChainVercel AI SDKMastra AILlamaIndexCrewAIPydantic AIAutoGen

What is Composio Connect?

Composio Connect lets anyone plug 1,000+ applications directly into their agent harness, including DeepSeek Harness, through a single hosted MCP endpoint. Each task runs as a session, a scoped runtime that ties together the user, the available toolkits, authentication, and execution state. It can:

  • Discover and load tools on-demand. A small set of meta tools lets the agent search for the right tool at runtime, so you never load hundreds of tool definitions into context.
  • Chain tools in a sandboxed remote workbench. Multi-step workflows run server-side, passing outputs between tools and working with files in the session's sandbox, so long chains finish without a lot of back-and-forth with the LLM.
  • Manage authentication end-to-end. Managed OAuth, per-user connected accounts, and account isolation are all handled for you, so you never wire up auth yourself.

Connect Tavily to DeepSeek Harness with Composio

Prerequisites: Install DeepSeek Harness

DeepSeek Harness needs Node.js (20 or 22) and pnpm. Build it from source. This is also what the MCP path in Option 2 requires, since the built-in MCP client resolves its dependencies from the checkout:

bash
git clone https://github.com/deepseek-ai/deepseek-harness.git
cd deepseek-harness
pnpm install
pnpm run build
pnpm dsh web

The Web UI listens on http://127.0.0.1:3080 by default. Open that address to confirm it is up.

Then configure your model route. In the Web UI, open Settings → Models: the built-in DeepSeek card accepts a DeepSeek API key, and the same page supports Anthropic, OpenAI, and custom OpenAI-compatible endpoints. Full first-run details are in the DeepSeek Harness docs.

Option 1: Composio CLI (recommended)

DeepSeek Harness runs shell commands as a first-class tool through its dsh-base bundle, so the Composio Universal CLI fits well. There is no MCP server to configure, and multi-step tool chaining is reliable. Install and authenticate:

bash
# Install the Composio CLI
curl -fsSL https://composio.dev/install | bash

# Log in to Composio
composio login

Connect to Tavily

Ask DeepSeek Harness to connect to Tavily, or just request any Tavily-related task. Under the hood the agent calls Composio commands like any other shell tool:

bash
composio search "what can I do with Tavily?"
composio link tavily

On the first Tavily call, composio link tavily opens an OAuth prompt so you can authorize access. After that, every Tavily command works with your stored credentials automatically, so you just talk to the agent and it will search, execute, and chain Tavily tools as needed.

Option 2: Composio MCP

DeepSeek Harness bridges MCP through its built-in @deepseek-ai/dsh-mcp-client plugin. You register one server as a plugin row, and its tools appear under server-qualified names (mcp__<serverName>__<rawName>), the same shape Claude Code and Codex use.

MCP requires the source build. The built-in client depends on the harness's own internal packages (cordis, dsh-llm, dsh-tools, dsh-invariants, dsh-subprocess), which resolve from the git checkout you built in the prerequisites. Run everything below from that checkout with pnpm dsh. For a route that avoids editing config files, see the community plugin at the end of this section.

Auth note: The built-in client passes static headers only and has no OAuth flow, so authenticate to Composio Connect with the x-consumer-api-key header. For Composio's OAuth flow, use the community plugin below.

1. Register the Composio server

The web profile's user patch layer is $DSH_HOME/profiles/web/cordis.patch.yml ($DSH_HOME defaults to ~/.dsh). This is a patch file, so a new plugin goes under an insert: entry. A bare - id: row instead targets an existing entry and fails to boot with patch: entry "…" not found. Add:

bash
- insert:
    - id: mcp-composio
      name: '@deepseek-ai/dsh-mcp-client'
      config:
        serverName: composio
        transport: streamable-http
        url: https://connect.composio.dev/mcp
        headers:
          x-consumer-api-key: !!js process.env.COMPOSIO_API_KEY

2. Set the API key and launch

The header value must resolve to a string at load time. Export your Composio consumer API key (ck_...), which is on the Connect → Settings → Sessions page of the Composio dashboard, in the shell you launch from. If it is unset, process.env.COMPOSIO_API_KEY is undefined and the config fails validation:

bash
export COMPOSIO_API_KEY=ck_your_key
pnpm dsh web

3. Done!

Open http://127.0.0.1:3080 and start a session on Standard or Code mode. (Minimal mode exposes only bash and str_replace_editor, so it hides MCP tools.) Composio's tools register as mcp__composio__*. Ask DeepSeek Harness to connect to Tavily or request any Tavily-related task, and Composio authorizes Tavily on demand on the first call.

The dsh-mcp-client config schema (transport, serverName, url, headers, plus reconnect.* and toolCallTimeoutMs) is the current developer-preview interface and may change between releases.

Alternative: OAuth via the community plugin

To use Composio Connect's OAuth flow instead of a static key, or to skip editing cordis.patch.yml by hand, use the community dsh-mcp-manager plugin. It installs as a self-contained external plugin, ships a bundle patch so it auto-mounts, and adds a Settings → MCP page plus OAuth (authorization code + PKCE, RFC 7591 dynamic client registration, and refresh-token rotation). It needs the web profile and Node.js ^22.19 or >=24, with pnpm on your PATH.

Install it from your source checkout, then restart pnpm dsh web and refresh:

bash
pnpm dsh plugin --profile web add github:hyqhyq3/dsh-mcp-manager

Open Settings → MCP → ＋ Add MCP server, choose HTTP, set the name to composio, the URL to https://connect.composio.dev/mcp, and the auth mode to OAuth. Click Authenticate, and Composio's login and consent screen opens in the browser, then redirects back and registers the tools as mcp__composio__*. The row should read connected (N tools).

The plugin catches the OAuth code on a loopback redirect (http://127.0.0.1:<port>/mcp-manager/callback/<id>), so Composio's OAuth provider must allow that loopback URL. Server configs and tokens persist at ~/.dsh/mcp-manager.json, so treat that file as a secret. Being a community plugin on a preview harness, its commands may change, so check the plugin repo if a step drifts.

What is the Tavily MCP server, and what's possible with it?

The Tavily MCP server is an implementation of the Model Context Protocol that connects your AI agent and assistants like Claude, Cursor, etc directly to your Tavily account. It provides structured and secure access to advanced web search and data retrieval, so your agent can perform actions like searching the web, filtering results, setting search parameters, and extracting relevant information from online sources on your behalf.

  • Custom web search with filters: Ask your agent to search the web with custom depth, result count, and domain restrictions for highly targeted results.
  • Document and content type discovery: Direct your agent to locate specific types of content—like articles, PDFs, or news—from across the internet.
  • Relevant data extraction: Have your agent pull and summarize key information from search results to save you time and effort.
  • Domain-specific research: Instruct your agent to confine searches to specific websites or sources for more trustworthy or relevant outcomes.
  • Efficient knowledge retrieval: Let your agent quickly surface facts, references, or recent developments from the web without manual browsing.

Way Forward

With Tavily connected, DeepSeek Harness can now act on your behalf whenever you ask it to.

From here, you can extend the harness further:

  • Connect more apps: Calendar, Slack, Notion, Linear, and hundreds of others are available through the same Composio Connect setup, and each new integration adds to what your agent can do for you.
  • Build workflows across tools: Once multiple apps are connected, DeepSeek Harness can chain actions together, so it can turn an email into a calendar invite, a Slack message into a Linear ticket, or a meeting note into a follow-up draft.
  • Use the plugin system: Wrap common Tavily flows as dsh plugins or prompt templates so recurring tasks become one-line commands, and bundle your most-used Tavily tools into a named profile you can boot on demand.

If you run into trouble or want to share what you've built, join the community or check out the Docs for more configuration options.

TOOLS

Supported Tools

Every Tavily action and event your agent gets out of the box.

Tavily crawl

Tool to perform intelligent graph-based website crawling with parallel path exploration and content extraction.

Tavily extract

Tool to extract and parse web page content from specified URLs using Tavily's extract endpoint.

Get Tavily usage

Tool to retrieve API key and account usage statistics from Tavily.

Tavily map website

Tool to map a website and discover its pages.

Tavily search

Use this to perform a web search via the Tavily API; offers controls for search depth, content types, result count, and domain filtering.

FRAMEWORKS

How to build Tavily MCP Agent with another framework

ChatGPT Work logo

ChatGPT Work

Use Tavily MCP with ChatGPT Work

Antigravity logo

Antigravity

Use Tavily MCP with Antigravity

OpenAI Agents SDK logo

OpenAI Agents SDK

Use Tavily MCP with OpenAI Agents SDK

Claude Agent SDK logo

Claude Agent SDK

Use Tavily MCP with Claude Agent SDK

Claude Code logo

Claude Code

Use Tavily MCP with Claude Code

Claude Cowork logo

Claude Cowork

Use Tavily MCP with Claude Cowork

Codex logo

Codex

Use Tavily MCP with Codex

Kimi Code logo

Kimi Code

Use Tavily MCP with Kimi Code

Grok Build logo

Grok Build

Use Tavily MCP with Grok Build

Cursor logo

Cursor

Use Tavily MCP with Cursor

VS Code logo

VS Code

Use Tavily MCP with VS Code

OpenCode logo

OpenCode

Use Tavily MCP with OpenCode

OpenClaw logo

OpenClaw

Use Tavily MCP with OpenClaw

Hermes logo

Hermes

Use Tavily MCP with Hermes

Pi logo

Pi

Use Tavily MCP with Pi

CLI logo

CLI

Use Tavily MCP with CLI

Google ADK logo

Google ADK

Use Tavily MCP with Google ADK

LangChain logo

LangChain

Use Tavily MCP with LangChain

Vercel AI SDK logo

Vercel AI SDK

Use Tavily MCP with Vercel AI SDK

Mastra AI logo

Mastra AI

Use Tavily MCP with Mastra AI

LlamaIndex logo

LlamaIndex

Use Tavily MCP with LlamaIndex

CrewAI logo

CrewAI

Use Tavily MCP with CrewAI

Pydantic AI logo

Pydantic AI

Use Tavily MCP with Pydantic AI

AutoGen logo

AutoGen

Use Tavily MCP with AutoGen

FAQ

Frequently asked questions

With a standalone Tavily MCP server, the agents and LLMs can only access a fixed set of Tavily tools tied to that server. However, with the Composio Tool Router, agents can dynamically load tools from Tavily and many other apps based on the task at hand, all through a single MCP endpoint.

Yes, you can. DeepSeek Harness fully supports MCP integration. You get structured tool calling, message history handling, and model orchestration while Tool Router takes care of discovering and serving the right Tavily tools.

Yes, absolutely. You can configure which Tavily scopes and actions are allowed when connecting your account to Composio. You can also bring your own OAuth credentials or API configuration so you keep full control over what the agent can do.

All sensitive data such as tokens, keys, and configuration is fully encrypted at rest and in transit. Composio is SOC 2 Type 2 compliant and follows strict security practices so your Tavily data and credentials are handled as safely as possible.

Start with Tavily.It takes 30 seconds.

Managed auth, hosted MCP servers, and every Tavily tool your agent needs.Free to start.

Start building