Skip to content

FREE_MCP_SERVERS.md

Latest commit

 

History

History
183 lines (140 loc) · 5.91 KB

FREE_MCP_SERVERS.md

File metadata and controls

183 lines (140 loc) · 5.91 KB
Raw

Free MCP Servers for Swarms Agents

A catalog of real, public MCP servers you can plug straight into the Swarms Agent class — most require no authentication, a few use a free-tier API key. Runnable examples live in this folder, numbered 01–05.

Availability and URLs change over time — verify a server before depending on it. Last reviewed against known-public servers as of early 2026.

Using an MCP server with an Agent

The Agent class connects to MCP servers over HTTP. Point it at a URL and it fetches that server's tools on startup; the model calls them as needed.

from swarms import Agent

agent = Agent(
    agent_name="MCP-Agent",
    model_name="gpt-4o-mini",              # any LiteLLM model with tool use
    mcp_url="https://mcp.deepwiki.com/mcp",  # one server
    max_loops=1,
)
print(agent.run("Use your tools to explain the kyegomez/swarms repo."))

Several servers at once — pass a list; the agent gets the union of their tools:

agent = Agent(
    model_name="gpt-4o-mini",
    mcp_urls=[
        "https://mcp.deepwiki.com/mcp",
        "https://learn.microsoft.com/api/mcp",
    ],
    max_loops=2,
)

Transport is auto-detected: any https:// URL uses streamable-HTTP. A few servers are SSE-only (endpoint ends in /sse) — most of those also expose a streamable-HTTP endpoint; prefer it when available.

Free — no authentication (works out of the box)

Server URL What it does
DeepWiki https://mcp.deepwiki.com/mcp Q&A over any public GitHub repo's docs
GitMCP https://gitmcp.io/<owner>/<repo> Turns a single repo into a docs/code MCP server
Microsoft Learn https://learn.microsoft.com/api/mcp Official Microsoft / Azure / .NET documentation
AWS Knowledge https://knowledge-mcp.global.api.aws Official AWS docs, API references, what's-new
Cloudflare Docs https://docs.mcp.cloudflare.com/mcp Cloudflare product documentation
Hugging Face https://huggingface.co/mcp Search models / datasets / Spaces (optional HF token unlocks more)
Context7 https://mcp.context7.com/mcp Up-to-date docs for thousands of libraries (rate-limited without a free key)
Globalping https://mcp.globalping.dev/sse Run ping / traceroute / DNS / MTR from a global probe network
Semgrep https://mcp.semgrep.ai/mcp Static-analysis security scanning of code 
# Example: repo-scoped documentation assistant via GitMCP
agent = Agent(
    model_name="gpt-4o-mini",
    mcp_url="https://gitmcp.io/kyegomez/swarms",
    max_loops=1,
)

Free tier — needs a free API key

These are free to use but require a key. How the key is passed differs per server.

Server Endpoint / auth What it does
Exa https://mcp.exa.ai/mcp?exaApiKey=… (query param) Web search + content retrieval
Tavily https://mcp.tavily.com/mcp/?tavilyApiKey=… (query param) Web search built for agents
Firecrawl hosted MCP + FIRECRAWL_API_KEY Scrape/crawl sites into clean markdown
Ref (ref.tools) hosted + key Fast documentation search across frameworks
Brave Search hosted/local + key Web + local search
Apify / Bright Data hosted + key Web-scraping / data-extraction actors

Key as a query parameter (Exa, Tavily):

import os
key = os.environ["EXA_API_KEY"]  # free at https://dashboard.exa.ai/api-keys
agent = Agent(
    model_name="gpt-4o-mini",
    mcp_url=f"https://mcp.exa.ai/mcp?exaApiKey={key}",
    max_loops=1,
)

Key as a Bearer token — the Agent sends Authorization: Bearer <key>; use env:VAR to keep the secret out of source:

agent = Agent(
    model_name="gpt-4o-mini",
    mcp_url="https://example-server.com/mcp",
    mcp_api_key="env:MY_SERVER_KEY",
)

Key in a custom header:

agent = Agent(
    model_name="gpt-4o-mini",
    mcp_url="https://example-server.com/mcp",
    mcp_headers={"x-api-key": os.environ["MY_SERVER_KEY"]},
)

OAuth — free with your own account (not no-auth)

These authenticate to your account and give the agent real actions in that service. Provide a token via mcp_api_key="env:VAR" (Bearer) or mcp_headers.

GitHub (https://api.githubcopilot.com/mcp/), Notion, Linear, Asana, Atlassian, Sentry, Stripe, PayPal, Vercel, Cloudflare (bindings), and many more.

agent = Agent(
    model_name="gpt-4o-mini",
    mcp_url="https://api.githubcopilot.com/mcp/",
    mcp_api_key="env:GITHUB_TOKEN",
)

Where to discover more

Registries that list hundreds of servers with their transports and auth:

Model note

Some current Anthropic models (e.g. claude-sonnet-5, claude-opus-4-8) can fail through the LiteLLM wrapper with:

AnthropicException: "thinking.type.enabled" is not supported for this model.
Use "thinking.type.adaptive" and "output_config.effort".

Until that's fixed in the wrapper, the examples use gpt-4o-mini, which works cleanly for MCP tool use. Swap MODEL for any LiteLLM-supported model you have a key for.

Runnable examples

The numbered examples in this folder:

File Server Auth
01_deepwiki_repo_qa.py DeepWiki none
02_gitmcp_repo_docs.py GitMCP none
03_microsoft_learn_docs.py Microsoft Learn none
04_multi_server_agent.py DeepWiki + Microsoft Learn none
05_exa_web_search.py Exa free API key 
export OPENAI_API_KEY=...        # or ANTHROPIC_API_KEY, etc.
python examples/mcp/agents/01_deepwiki_repo_qa.py