Skip to the page
Get started free

DuckDuckGo MCP server

Setup, limits and search at $1 per 1,000.

The DuckDuckGo MCP server on PyPI installs with uvx duckduckgo-mcp-server, needs no key and gives your agent search and fetch_content. It holds itself to 30 searches a minute, and DuckDuckGo blocks its default client unless you add the [browser] extra.

SpecStream is the live web data engine for teams and agents. Its hosted MCP server, https://api.specstream.ai/mcp, gives your agent the most accurate web search API in the world at $1 per 1,000 searches, with nothing to install and $10 of free credit.

Get an API key

New accounts get $10 of credit, no card. It covers 10,000 fast searches.

Three DuckDuckGo MCP servers and two hosted servers, side by side

SpecStreamduckduckgo-mcp-server (Python)duckduckgo-mcp-server (npm)Web ForagerExa MCP
Install line or addresshttps://api.specstream.ai/mcp over Streamable HTTP, with OAuth or an API keyuvx duckduckgo-mcp-server, or claude mcp add ddg-search uvx duckduckgo-mcp-server in Claude Codenpx -y duckduckgo-mcp-server in the client's configuvx --python ">=3.10,<3.14" web-forager servehttps://mcp.exa.ai/mcp
Key and accountA free account, no card. A client that signs in with OAuth opens the browser once, and no key sits in a fileNone in the configNone in the configNone in the configNone at the free rate limit. From IPv6 or a proxied network, sign in with OAuth or add a key
Where the search runsOn SpecStream's servers, across several search engines. Nothing runs on your machineOn your machine. DuckDuckGo blocks the default client with an empty HTTP 202 page, and search needs the [browser] extra to passOn your machineOn your machine. Web search falls back to Brave when DuckDuckGo failsOn Exa's servers
LimitsThe spending limit you set on each key30 searches and 20 page fetches a minute by default1 request a second and 15,000 a monthNo retry loop or rate limiting of its ownA free rate limit
Tools your agent getssearch_web, read_pages, and Deep Search through compile, quote_capacity_work and authorize_capacity_worksearch, fetch_content and expand_linkduckduckgo_search, 1 to 20 results a callduckduckgo_search, duckduckgo_news_search and web_fetchSearch and fetch. Exa Agent needs sign in or a key
What a search returnsRanked sources with title, URL and snippet. The default search adds the answer with the page it was read from and the read timeTitles, URLs and snippets as formatted textMarkdown search resultsJSON by default, or textSearch results, and page text through fetch
What it costs$1 per 1,000 searches of up to 10 results on the fast search, $5 with the answer, $0.80 per 1,000 pages read. $10 of credit to start, no cardFree software under the MIT LicenseFree to runFree to runFree at the rate limit. Past it, from $4 per 1,000 requests

Why agents search through SpecStream

No empty results when DuckDuckGo blocks

A DuckDuckGo server reads DuckDuckGo's results page from your machine. DuckDuckGo now answers its default client with an empty HTTP 202 page, which the server reads as no results, and the fix is a second install, the [browser] extra. A SpecStream search runs on SpecStream's servers across several search engines, and a search that returns nothing costs nothing.

One line in Claude Code

Run this and sign in to SpecStream when the browser opens. There is no Python, no Node.js and no browser extra to install:

claude mcp add --transport http specstream https://api.specstream.ai/mcp

Cursor, VS Code, ChatGPT and the other clients take the same address. The MCP client guides show where each one keeps it.

The most accurate web search API in the world, as one tool call

search_web runs SpecStream's Search API. It scores 97.9 F-score on all 1,000 official SimpleQA Verified questions (978 correct), and 64.0 item F1 and 43.4 row F1 on all 200 official WideSearch tasks. The Extract API behind read_pages scores 0.908 F1 on WCXB. The scores and the method are on the benchmarks page.

A results page hands your agent ten links to open. The default search hands it the answer, with the page it was read from and the read time, when the pages agree on a value, and null when they do not.

Whole pages, PDFs included

fetch_content returns 8,000 characters a call by default and pages through the rest. read_pages reads up to 20 URLs a call, PDF, Word and Excel files included, and returns each page's main text as Markdown with its title and the date it says it was published, at $0.80 per 1,000 pages. A page with no text is not charged.

Search operators hold on every result

site:, -site:, filetype:, quoted phrases and after: or before: hold on every result SpecStream returns. A result that breaks one is left out, and SpecStream searches the engines' next pages to fill the list.

Your agent asks for data and gets rows

When the job is a list, the same connection runs Deep Search. compile reads the request into conditions and facts you can edit, quote_capacity_work prices it, and authorize_capacity_work runs it under the maximum you approve. Rows cost 5¢, 15¢ or $1 each, with the source URL and read time on every value, and a run that finds nothing costs nothing.

Other DuckDuckGo MCP alternatives

Docker's mcp/duckduckgo image

Docker's MCP catalog carries mcp/duckduckgo, which its page calls a community maintained server, not published by or affiliated with DuckDuckGo. It runs in a container on your machine. SpecStream's server is hosted, so there is no container to run.

Parallel MCP

Parallel's hosted server at https://mcp.parallel.ai/mcp needs no account and no key, and web_search, web_fetch and web_decisions are free at lower rate limits. SpecStream's free start is $10 of credit with no daily allowance, spent at $1 per 1,000 fast searches.

Brave Search MCP

Brave's official server runs on your machine and needs a BRAVE_API_KEY. The Search API costs $5 per 1,000 requests and includes $5 in free credits every month. SpecStream's fast search is $1 per 1,000, with page reading on the same server.

SearXNG MCP

mcp-searxng searches through a SearXNG instance you name in SEARXNG_URL, and the instance must answer in JSON: SearXNG's docs say an unset format returns 403 Forbidden and that many public instances have it turned off. SpecStream answers SearXNG's own address, so a tool built for SearXNG works with no server to run. See moving off SearXNG.

How SpecStream scores

The same numbers as the benchmarks page, from the official tasks and scorers.

SpecStreamParallelExaTavilyPerplexity
Web researchWideSearch, GPT-5.6 Luna60.945.553.047.947.0
Factual accuracySimpleQA Verified, GPT-5.6 Luna9694919494
Factual accuracySimpleQA Verified, GPT-5.6 Sol9997919295

How to move from a DuckDuckGo server to SpecStream

1. Sign up at specstream.ai. The account starts with $10 of credit, no card.

2. In Claude Code, remove the DuckDuckGo server and add SpecStream, then sign in when the browser opens:

claude mcp remove ddg-search
claude mcp add --transport http specstream https://api.specstream.ai/mcp

3. In Claude Desktop, delete the ddg-search entry from claude_desktop_config.json and add https://api.specstream.ai/mcp as a custom connector.

4. In Cursor, create a key under Agents & API in the workspace, put it in SPECSTREAM_API_KEY, and replace the DuckDuckGo entry in mcp.json with this:

{
  "mcpServers": {
    "specstream": {
      "url": "https://api.specstream.ai/mcp",
      "headers": { "Authorization": "Bearer ${env:SPECSTREAM_API_KEY}" }
    }
  }
}

5. Your agent calls search_web where it called search, and read_pages where it called fetch_content.

Get an API key

Frequently asked questions

Something we didn't answer? Talk to a human

Is there an official DuckDuckGo MCP server?
The DuckDuckGo MCP servers are community projects: Docker's mcp/duckduckgo image says it is not published by or affiliated with DuckDuckGo. The one on PyPI, duckduckgo-mcp-server, is MIT licensed.
How do I install the DuckDuckGo MCP server?
Run uvx duckduckgo-mcp-server, or claude mcp add ddg-search uvx duckduckgo-mcp-server in Claude Code. Use uvx --with "duckduckgo-mcp-server[browser]" duckduckgo-mcp-server so search still works when DuckDuckGo blocks the default client.
Why does the DuckDuckGo MCP server return no results?
DuckDuckGo answers the server's default client with an empty HTTP 202 page, which reads as no results. Search works only with the server's [browser] extra installed, and without it search returns a message asking you to install it.
Does the DuckDuckGo MCP server need an API key?
No. It needs no key and no account, and runs on your machine at up to 30 searches a minute by default. SpecStream's hosted server needs a free account, and a client that signs in with OAuth needs no key.
What does SpecStream's MCP search cost?
$1 per 1,000 searches of up to 10 results on the fast search, $5 per 1,000 with the answer, and $0.80 per 1,000 pages read. New accounts get $10 of credit with no card.

Sources

SpecStream wrote this page. Each install line, limit and price comes from the project's own README or the company's own page, read on Nov 12, 2026.

  1. duckduckgo-mcp-server READMEgithub.com/nickclyde/duckduckgo-mcp-server, read Nov 12, 2026
  2. duckduckgo-mcp-server on PyPIpypi.org/project/duckduckgo-mcp-server/, read Nov 12, 2026
  3. duckduckgo-mcp-server (npm) READMEgithub.com/zhsama/duckduckgo-mcp-server, read Nov 12, 2026
  4. Web Forager READMEgithub.com/CyranoB/web-forager, read Nov 12, 2026
  5. Docker mcp/duckduckgohub.docker.com/r/mcp/duckduckgo, read Nov 12, 2026
  6. Exa MCPdocs.exa.ai/reference/exa-mcp, read Nov 12, 2026
  7. Exa pricingexa.ai/pricing, read Nov 12, 2026
  8. Parallel MCPparallel.ai/mcp, read Nov 12, 2026
  9. Brave Search MCP servergithub.com/brave/brave-search-mcp-server, read Nov 12, 2026
  10. Brave Search APIbrave.com/search/api/, read Nov 12, 2026
  11. mcp-searxng READMEgithub.com/ihor-sokoliuk/mcp-searxng, read Nov 12, 2026
  12. SearXNG search APIdocs.searxng.org/dev/search_api.html, read Nov 12, 2026
  13. SpecStream benchmarksspecstream.ai/benchmarks, read Nov 12, 2026
  14. SpecStream for agentsspecstream.ai/agents.md, read Nov 12, 2026