GitHub - AnswerDotAI/solvemcp: Small Python client for Model Context Protocol (MCP) servers.

It provides:

  • stdio transport for local subprocess servers.
  • Streamable HTTP transport for modern MCP servers.
  • Legacy HTTP+SSE fallback for older servers.
  • Dynamic Python callables for tools returned by tools/list.

Installation

Quick Start

Connect to a local stdio server

from solvemcp import MCPClient
with MCPClient.stdio([sys.executable, "-u", "../tests/echoserver.py"]) as mcp:
    print(list(mcp.tools))
    res = mcp.echo(text="hello")
    print(res)
['echo']
{'content': [{'type': 'text', 'text': 'hello'}]}

Connect to an HTTP MCP server

with MCPClient.http("https://mcp.grep.app") as mcp:
    print(list(mcp.tools))
    res = mcp.searchGitHub(query="class PtyProcess", language=["Python"])
mcp = MCPClient.http("https://mcp.grep.app")
print(list(mcp.tools))
res = mcp.searchGitHub(query="class PtyProcess", language=["Python"])
def searchGitHub(
    query:Any, matchCase:Any=False, matchWholeWords:Any=False, useRegexp:Any=False, repo:Any=None, path:Any=None,
    language:list=None
):
Find real-world code examples from over a million public GitHub repositories to help answer programming questions.

**IMPORTANT: This tool searches for literal code patterns (like grep), not keywords. Search for actual code that would appear in files:**
- ✅ Good: 'useState(', 'import React from', 'async function', '(?s)try {.*await'
- ❌ Bad: 'react tutorial', 'best practices', 'how to use'

**When to use this tool:**
- When implementing unfamiliar APIs or libraries and need to see real usage patterns
- When unsure about correct syntax, parameters, or configuration for a specific library
- When looking for production-ready examples and best practices for implementation
- When needing to understand how different libraries or frameworks work together

**Perfect for questions like:**
- "How do developers handle authentication in Next.js apps?" → Search: 'getServerSession' with language=['TypeScript', 'TSX']
- "What are common React error boundary patterns?" → Search: 'ErrorBoundary' with language=['TSX']
- "Show me real useEffect cleanup examples" → Search: '(?s)useEffect\(\(\) => {.*removeEventListener' with useRegexp=true
- "How do developers handle CORS in Flask applications?" → Search: 'CORS(' with matchCase=true and language=['Python']

Use regular expressions with useRegexp=true for flexible patterns like '(?s)useState\(.*loading' to find useState hooks with loading-related variables. Prefix the pattern with '(?s)' to match across multiple lines.

Filter by language, repository, or file path to narrow results.

File: ~/aai-ws/toolslm/toolslm/funccall.py

Type: function

cts = res['content']
print(len(cts))
print(cts[0]['text'][:500])
Repository: sedwards2009/extraterm
Path: extensions/ProxySessionBackend/src/python/ptyprocess/ptyprocess.py
URL: https://github.com/sedwards2009/extraterm/blob/master/extensions/ProxySessionBackend/src/python/ptyprocess/ptyprocess.py
License: MIT

Snippets:
--- Snippet 1 (Line 79) ---
            (intr, eof) = (3, 4)
    
    _INTR = _byte(intr)
    _EOF = _byte(eof)

class PtyProcessError(Exception):
    """Generic error class for this package."""

# setecho and setwinsize are pulled out here b

How Tool Calls Work

For each server tool with a valid Python identifier name, MCPClient adds a method dynamically. Example: a server tool named echo becomes mcp.echo(...).

Equivalent forms:

mcp.echo(text="hi")
mcp.call_tool("echo", text="hi")

Streaming

For Streamable HTTP servers, use rpc_stream or call_tool_stream:

for msg in mcp.rpc_stream("tools/list", params={}): print(msg)

Development

Module Layout

  • solvemcp/client.py: MCPClient request lifecycle, initialization, tool binding, and RPC helpers.
  • solvemcp/transports.py: transport implementations plus SSE parsers and error types.

Run tests: