Skip to main content

MCP Tools

The Model Context Protocol (MCP) enables AI agents to access external tools and integrations. Strongly AI ships with a catalog of more than 130 pre-built MCP servers covering databases, APIs, productivity tools, and more, plus support for your own custom MCP servers.

How It Works​

Each organization has a single MCP tool hub that serves MCP servers' tools to your agents and workflows. Each user activates a server with their own configuration:

  1. You enter an MCP server's configuration (your API keys, URLs) and activate it. Its tools are registered with your organization's tool hub for you alone
  2. Your enabled tools become available to your agents and workflows through the MCP Tools Provider node
  3. During a run, the agent decides which tools to call, and each call uses the configuration of the user the run belongs to: whoever started the agent or deployed the workflow. A teammate's run never uses your keys, and yours never uses theirs
  4. Registrations persist: your enabled MCPs survive platform restarts

Available MCP Servers​

Search & Web​

  • Brave Search -- web, image, video, news search
  • DuckDuckGo -- privacy-focused search (no API key)
  • Exa -- AI-powered semantic search
  • Firecrawl -- web scraping and data extraction
  • Google Translate -- translation for 100+ languages

Development & DevOps​

  • GitHub -- repos, issues, PRs, code search (26 tools)
  • Git -- repository operations
  • Docker Hub -- image search, tags, repositories
  • SonarQube -- code quality analysis
  • Postman -- API testing and management
  • CircleCI -- CI/CD pipelines
  • Netlify / Render -- deployment platforms

Databases​

  • Redis -- key-value, hashes, lists, sets
  • Elasticsearch -- full-text search and analytics
  • MongoDB -- document CRUD, aggregation
  • Neo4j -- graph database Cypher queries
  • SingleStore -- distributed SQL
  • Couchbase -- NoSQL document database
  • CockroachDB -- distributed SQL

Productivity​

  • Atlassian -- Jira, Confluence, Bitbucket (16 tools)
  • Notion -- pages, databases, search
  • Todoist -- task management
  • Google Tasks / Microsoft To Do -- task lists

Communication​

  • Mailgun -- transactional email
  • LinkedIn -- post content, profile
  • Webex / Mattermost / RocketChat / Matrix -- messaging

Finance & Commerce​

  • Stripe -- payments, customers, subscriptions
  • Coinbase -- cryptocurrency data
  • Razorpay -- payment processing
  • Xero -- accounting

AI & Data​

  • Wolfram Alpha -- computational knowledge
  • DeepL -- neural machine translation
  • ElevenLabs -- text-to-speech
  • OpenWeather -- weather data and forecasts

Monitoring​

  • Grafana -- dashboards, alerts, data sources
  • Dynatrace -- APM monitoring

Utilities​

  • HTTP Fetch -- generic web requests (no API key)
  • RSS -- feed parsing (no API key)
  • Time -- timezone operations
  • Playwright -- browser automation

...and many more. Browse the full catalog on the Workflow Tools page (/workflow-tools).

Configuring an MCP Server​

  1. Navigate to the Workflow Tools page (/workflow-tools). It lists all MCP servers and custom tools, with counts and a type filter
  2. Open a server's Details page and enter your configuration values (API keys, URLs) on its Configuration tab. They are never saved on the shared server entry
  3. Click Activate. This registers the server for you with your values; the page then shows the server as Active for you, with when you registered it. Each user sees only their own state: a server a teammate activated shows as Not active until you activate it with your own values
  4. Use Disable / Enable to switch your registration off or on, Re-register to activate again with new values, and Deactivate to remove your registration
  5. Use the server's tools in workflows via the MCP Tools Provider node

Calling a tool of a server you have not activated fails with a message telling you to activate it with your own configuration first.

The card menu also supports Clone (copy a server so you can vary its configuration) and Delete.

Using MCP Tools in a Workflow​

The MCP Tools Provider node (Operators category, node type mcp-tools-provider) supplies MCP tools to agent nodes:

  1. Drag an MCP Tools Provider node onto the canvas
  2. In its configuration, select the MCP server from the dropdown of deployed servers (each entry shows its Running / Not Deployed status). An additional multi-select lets you attach more servers to the same provider
  3. Connect the provider's output to the agent node's tools connector (bottom of the agent node)
  4. The agent autonomously calls the provided tools as needed during a run

The node outputs the aggregated tool list (tools), the number of servers queried (serverCount), and the total tool count (toolCount). When you deploy the workflow, the platform resolves the selected servers, registers them with your organization's tool hub, and populates the provider with the live tool definitions automatically.

Adding Custom MCP Servers​

You can add your own MCP server from the Workflow Tools page:

  1. Click to create a new tool and choose the MCP server type
  2. Upload a ZIP or TAR bundle containing mcp.py at its root (requirements.txt and README.md are optional)
  3. Save, then Activate it with your configuration values. Uploading new code later requires a version bump

The tool hub runs exactly the bundle you published: it checks the downloaded files against the digest recorded when you uploaded them and refuses to run anything else. A custom server is listed in the workflow builder once its bundle is published.

mcp.py must declare its tools and an execute entry point:

# mcp.py
import httpx
from typing import Dict, Any

TOOLS = [
    {"name": "my_tool", "description": "Does something useful",
     "inputSchema": {"type": "object", "properties": {"query": {"type": "string"}}, "required": ["query"]}},
]

async def execute(tool_name: str, arguments: Dict[str, Any], config: Dict[str, str]) -> Dict[str, Any]:
    async with httpx.AsyncClient() as c:
        r = await c.get("https://api.example.com/search", params={"q": arguments["query"]})
        return {"content": [{"type": "text", "text": r.text}]}
  • TOOLS is the list of tool definitions (name, description, JSON Schema input)
  • execute(tool_name, arguments, config) handles every tool call; config carries the calling user's own configuration values, the ones they activated the server with
  • Your code runs in your organization's tool hub, in its own sandbox, one process per call. It receives only arguments and config
  • Your code acts as no one. It carries no Strongly AI identity, credentials or keys, so it cannot call Strongly AI services: not the AI Gateway, the platform API, your organization's tool hub or the configuration of its other MCP servers. Its network reaches the public internet only; addresses inside the cluster, your private network and cloud metadata endpoints are refused. To call an external API, put that API's key in the server's configuration and read it from config
  • Packages in requirements.txt are installed into the sandbox the first time one of the server's tools is called
  • A bundle without mcp.py, or an mcp.py without a TOOLS = [...] declaration, is rejected at upload/deploy time