Ecosystem & Interop

🌐 120+ native tools

Ecosystem & Interoperability

The MCP AI Suite connects to the entire AI tooling ecosystem — no vendor lock-in.

Unlike frameworks that lock you into their ecosystem, the MCP AI Suite speaks three languages:

Dynamic MCP Client

Connect to any MCP server at runtime — no configuration, no restart. The engine discovers tools automatically and makes them available to the LLM.

from kernelmcp import KernelFactory

kernel = KernelFactory.create()

# Connect to community MCP servers
await kernel.orchestrator.connect_mcp_server(
    "github", transport="stdio", command="github-mcp-server"
)
await kernel.orchestrator.connect_mcp_server(
    "slack", transport="sse", url="http://localhost:8080/sse"
)

# Tools from GitHub and Slack are now available to the engine
task = await kernel.run("Create a GitHub issue for the bug reported in Slack #engineering")
# Engine automatically uses github__create_issue and slack__search_messages

The LLM can also connect servers itself

The engine exposes connect_mcp_server as a tool — the LLM can discover and connect to MCP servers autonomously:

User: "Connect to the Notion MCP server and list my pages"
Engine: → connect_mcp_server(name="notion", transport="stdio", command="notion-mcp-server")
        → notion__list_pages({})
        → "You have 42 pages. Here are the recent ones..."

Supported MCP Servers

Any server listed on mcp.so works out of the box:

• GitHub — Issues, PRs, repos, code search
• Slack — Messages, channels, users
• Notion — Pages, databases, blocks
• PostgreSQL — Query, schema, migrations
• Stripe — Payments, customers, invoices
• Jira — Issues, sprints, projects
• Linear — Issues, projects, teams
• Filesystem — Read, write, search files
• Google Drive — Documents, sheets, files
• And 2000+ more…

LangChain Tool Bridge

Use any of LangChain’s 500+ tools as native kernelmcp tools. Zero modification needed — just register and use.

from langchain_community.tools import WikipediaQueryRun, DuckDuckGoSearchRun
from kernelmcp import KernelFactory

kernel = KernelFactory.create()

# Register LangChain tools
kernel.orchestrator.register_langchain_tool(WikipediaQueryRun())
kernel.orchestrator.register_langchain_tool(DuckDuckGoSearchRun())

# Now the engine can use them
task = await kernel.run("Research quantum computing using Wikipedia and web search")

How it works

1. LangChain BaseTool is wrapped into an MCP-compatible tool schema
2. The orchestrator routes calls to the LangChain bridge transparently
3. Async tools run natively; sync tools are wrapped in asyncio.run_in_executor
4. Input schemas are auto-extracted from Pydantic args_schema when available

Compatible tools include:

Wikipedia, DuckDuckGo, Google Search, Arxiv, PubMed, Wolfram Alpha, Python REPL, Shell, SQL Database, OpenWeatherMap, Requests, JSON, Math, Human Input, File Management, and 500+ more from langchain-community.

LangChain Loader Bridge (RAG)

Ingest documents from any LangChain document loader into ragmcp’s vector pipeline. This gives ragmcp access to 80+ document formats and sources beyond its 15 native loaders.

from langchain_community.document_loaders import (
    PyPDFLoader, NotionDBLoader, S3FileLoader, GitLoader
)
from ragmcp import RAGFactory
from ragmcp.loaders.langchain_bridge import LangChainLoaderBridge

rag = RAGFactory.create_default()
bridge = LangChainLoaderBridge(rag)

# Ingest from any LangChain loader
await bridge.ingest_loader(PyPDFLoader("report.pdf"))
await bridge.ingest_loader(NotionDBLoader(token="secret_xxx"))
await bridge.ingest_loader(S3FileLoader(bucket="my-docs", key="data.csv"))
await bridge.ingest_loader(GitLoader(repo_path="./my-repo", branch="main"))

# All documents are chunked, embedded, and searchable in ragmcp
results = await rag.search("quarterly revenue analysis")

Supported formats (95+)

Documents: PDF, Word (.docx), Excel (.xlsx), PowerPoint (.pptx), CSV, JSON, HTML, Markdown, Plain Text, EPub, Email (.eml)

Cloud services: Notion, Confluence, Google Drive, S3, GCS, Slack, Jira, GitHub, GitLab

Media: YouTube transcripts, Audio (Whisper), Video (FFmpeg)

Data: SQL databases, HuggingFace datasets, Arxiv papers, PubMed, Wikipedia

Architecture: How Interop Works

Key design principle: Every external tool goes through the same execute_tool() pipeline — circuit breaker, audit logging, budget tracking, and observability apply uniformly to MCP servers, LangChain tools, and native tools alike.

Comparison with Other Frameworks