ProvenMapSCALE 1:1
New: architecture for every role
Docs
Pricing
Blog
Boards
Getting StartedCore ConceptsPlatform FeaturesReference & ResourcesSelf-HostedSources

Platform Features

Boards GuideArchetypesConnecting SourcesInsight SkillsSpecsIntentsPublishing BoardsSkills & RecipesChatMCP Servers
Platform Features
Boards GuideArchetypesConnecting SourcesInsight SkillsSpecsIntentsPublishing BoardsSkills & RecipesChatMCP Servers
Platform FeaturesMCP Servers

Last updated July 24, 2026

MCP Servers

The Model Context Protocol (MCP) is how you extend the platform beyond its built-in capabilities. Register an MCP server, and its tools, resources, and prompts become available to chat — so a conversation can reach databases, APIs, monitoring systems, or anything else you expose through the protocol.

What's an MCP server?

An MCP server is an external process that exposes capabilities through a standardized protocol. It can run locally (via stdio) or remotely (via SSE). Each server declares what it can do:

CapabilityDescription
ToolsFunctions chat can call — query a database, fetch metrics, run a script
ResourcesData chat can read — files, API responses, live system state
PromptsTemplated instructions chat can use for specific tasks
Tip

Every server you register widens what a conversation can reach — one for your metrics stack, one for your ticketing system, and chat can answer questions that span both.

Transport modes

MCP servers connect over one of two transports — stdio for a server running on the same machine, SSE for one running somewhere else. Streamable HTTP isn't supported yet; point a remote server at the SSE endpoint.

The server runs as a local process. You specify a command and optional arguments. Best for development, local tools, and servers that need filesystem access.

JSON
{
  "type": "stdio",
  "command": "npx",
  "args": ["-y", "@modelcontextprotocol/server-postgres"],
  "env": {
    "DATABASE_URL": "postgresql://..."
  }
}

When to use: Local development, database access, filesystem-based tools, anything that needs to run on the same machine.

Registering a server

Register your first MCP server

  • Go to Settings → MCP Servers in your org dashboard
  • Click Add Server
  • Enter a name and description for the server
  • Choose transport type — stdio for local, SSE for remote
  • Fill in the transport config (command/args for stdio, URL/headers for SSE)
  • Toggle which capability types are enabled (tools, resources, prompts)
  • Add categories and tags for organization
  • Click Save and verify the connection test passes

When you save, we validate the config, test the connection, and discover the server's capabilities straight away — so you see the test result before you leave the page.

Testing connections

After registering (or anytime), you can verify a server is reachable:

  1. Find the server in Settings → MCP Servers
  2. Click Test Connection
  3. The result shows success or failure with an optional diagnostic message

The server detail view tracks last-connected time and current status so you can monitor health at a glance.

Discovering capabilities

Once a server is connected, you can browse its capabilities:

  1. Navigate to the server's detail page
  2. The Capabilities tab lists all discovered tools, resources, and prompts
  3. Filter by type to narrow the list
  4. Click Refetch to re-discover capabilities if the server has been updated
What's captured per capability

Tools

Each discovered tool includes:

  • Name and LLM-friendly name — how chat references it
  • Input schema — structured definition of required parameters
  • Response type information — what the tool returns and how reliable the output schema is
  • Categories and usage examples — for discoverability
  • Enabled toggle — turn tools on or off without removing the server

Resources

  • URI and optional URI template for parameterized access
  • MIME type — what format the data is in
  • Subscribable flag — whether the resource supports real-time updates

Prompts

  • Arguments — named parameters with descriptions and required flags
  • Template — the prompt text
  • Tags and usage count — for organization and analytics
Note

Discovered capabilities go into the platform's tool catalog, and chat picks from that catalog each turn — so a well-described tool gets chosen more often than a vaguely named one. Names and descriptions are worth writing carefully.

Making tools available to chat

Once a server is registered and its capabilities discovered, its tools are available to chat. Chat resolves which ones are relevant to each conversation from:

  • The board's context and connected sources
  • The user's tool preferences
  • The server's capability categories

You don't need to manually wire tools to conversations — chat discovers and selects them dynamically.

Managing servers

  • Update — modify server config, transport settings, or capabilities
  • Delete — uninstall a server and remove its capabilities from the catalog
  • Search and filter — find servers by name or tags when you have many registered
Warning

Deleting an MCP server removes all its discovered capabilities from the platform. Any chat conversations that relied on those tools will no longer have access to them.

What's next

Chat

See how chat uses MCP tools in conversations.

Insight Skills

MCP tools can complement insight analysis.

Connecting Sources

MCP servers can act as data sources for your architecture.

PreviousChatNextIntegrations Reference

On this page

MCP ServersWhat's an MCP server?Transport modesRegistering a serverRegister your first MCP serverTesting connectionsDiscovering capabilitiesToolsResourcesPromptsMaking tools available to chatManaging serversWhat's next