Strands Agents
Give a Strands agent SocialScrape's tools through the hosted MCP server, in a few lines of Python, and pick which tools it sees.
Strands Agents agents take MCP servers as tools, so the hosted SocialScrape MCP server at https://api.socialscrape.dev/mcp gives a Strands agent the whole API: profiles, posts, comments, search and video transcripts across every platform, with nothing to install beyond Strands itself. The tools are the same ones the MCP guide describes.
Install
pip install -U "strands-agents>=1.65,<2" export SOCIALSCRAPE_API_KEY="YOUR_KEY"
Create a key in the dashboard; a new account starts with 100 free credits. The MCP client ships with Strands, so there is no separate package.
Connect an agent
import os
from strands import Agent
from strands.tools.mcp import MCPClient
socialscrape = MCPClient(
url="https://api.socialscrape.dev/mcp",
headers={"x-api-key": os.environ["SOCIALSCRAPE_API_KEY"]},
)
agent = Agent(tools=[socialscrape])
agent("How many followers does the TikTok account @nasa have?")
The agent opens the connection when it starts and closes it when it is done. Agent() uses Strands' default model provider, Amazon Bedrock. Any provider that supports tool calling works; for Anthropic's API, install strands-agents[anthropic] and pass a model:
from strands.models.anthropic import AnthropicModel
agent = Agent(
model=AnthropicModel(model_id="claude-haiku-5-5", max_tokens=1024),
tools=[socialscrape],
)
Choose the tools
The server has a tool per platform plus the tools agents use most. To keep the agent's tool list short, load only the ones you need with tool_filters, and add a prefix if the agent has tools from other servers too:
socialscrape = MCPClient(
url="https://api.socialscrape.dev/mcp",
headers={"x-api-key": os.environ["SOCIALSCRAPE_API_KEY"]},
tool_filters={"allowed": [
"get_profile", "get_post", "get_comments", "get_transcript", "search", "describe",
]},
prefix="socialscrape",
)
The agent then sees socialscrape_get_profile, socialscrape_get_post and so on.
| Tool | What it does |
|---|---|
get_profile | A profile, channel or page: name, bio, links and follower counts. The platform is an argument. |
get_post | One post or video from its URL. |
get_comments | The comments on a post or video, one page per call. |
get_transcript | The transcript of a video from its URL, from captions or AI transcription. |
search | Keyword search on one platform, one page per call. |
describe | The exact parameters, allowed values and price of any endpoint. Free, makes no request. |
The full list, including one tool per platform, call_endpoint for any endpoint by id and get_job for long transcriptions, is in the MCP guide.
Call a tool without a model
For a script or a test, open the client and call a tool directly:
with socialscrape:
result = socialscrape.call_tool_sync(
tool_use_id="1", name="get_credits", arguments={},
)
print(result["content"][0]["text"])
Every result starts with a line such as ok | credits charged: 1 | credits remaining: 4999. A request the API refuses comes back as an error result with the error code, so the agent can read it and try something else; see errors.
Credits
Each tool call that reaches the API is charged like any API request: see credits. Failures are free, describe makes no request, and max_age_days accepts a cached result for 0 credits, as described in caching.
Other agent frameworks
For LangChain and LangGraph, see the LangChain guide. For CrewAI, see the CrewAI guide. For Claude, ChatGPT, Cursor and other MCP clients, use the MCP server directly.