Skip to main content

​
What is MCP?

The Model Context Protocol (MCP) is an open standard that lets AI applications connect to external data sources and tools through a unified interface. Praxis AI Middleware includes a built-in MCP server, giving any MCP-compatible client — such as Claude Desktop, Cursor, or your own application — direct access to your Digital Twin’s knowledge base and API documentation.

Search Your Knowledge

Query your Digital Twin’s document collection using natural-language search via the search-instance-rag tool.

Explore the API

Access live OpenAPI documentation for both Runtime and Admin endpoints directly from your LLM.

Link Digital Twins

Connect multiple Digital Twins so one can query another’s knowledge base without duplicating data.

​
Quick Start

Follow these steps to expose your Digital Twin as an MCP server and connect a client.

​
Step 1 — Enable MCP on Your Digital Twin

Open your Digital Twin’s Configuration and Integrations panel and turn on the MCP Server switch, then save. Enable MCP Server switch

​
Step 2 — Copy Your MCP Secret

In the same panel, copy the value from the MCP Secret field. This token authenticates clients connecting to your MCP server. MCP Secret field location

​
Step 3 — Configure Your MCP Client

Use the following settings in any MCP-compatible client:
SettingValue
Server URLhttps://pria.praxislxp.com/api/mcp
Server Labelpria-mcp
AuthorizationBearer <your-mcp-secret>
Description / System Instruction (optional but recommended): 
Use this tool to run authoritative semantic search against the remote Digital Twin via the MCP server. Invoke search-instance-rag whenever the user's question targets content stored in the Digital Twin (prefer it over local search for remote-scope queries). Input is a natural-language query string; output is a relevance-ordered list of matches containing each item's file title, a concise summary, and a direct URL (include source_id and score when available). If no relevant results exist, return an empty list. Do not infer, synthesize, or modify content — return only what the remote instance provides.
The description tells the AI when and how to use the MCP tools. Customize it to match your use case.

​
Available Tools

Once connected, your MCP client can invoke these tools:
ToolDescription
search-instance-ragSemantic search across your Digital Twin’s uploaded documents and knowledge base
api-admin-docsReturns the Admin API OpenAPI specification for the Praxis AI Middleware
api-rt-docsReturns the Runtime API OpenAPI specification for the Praxis AI Middleware
Additional tools may be available depending on your Digital Twin’s configuration and enabled connectors.

​
Testing Your MCP Server

Use the MCP Inspector (documentation) to validate your connection, browse available tools, and send test requests. 
npx @modelcontextprotocol/inspector
Connect to https://pria.praxislxp.com/api/mcp with your Bearer token and verify that the tools listed above appear. MCP Inspector showing available tools

​
Linking Digital Twins via Connector MCP

A common use case is connecting two Digital Twins so that Twin B can search Twin A’s knowledge base in real time — without copying or syncing documents between them. Why link twins?
  • Keep source knowledge centralized in one twin while others query it on demand
  • Updates to Twin A’s documents are instantly available to Twin B
  • Avoid data duplication and keep all twins consistent

​
How to Set It Up

1

Enable MCP on Twin A

Follow the Quick Start above to enable MCP and copy Twin A’s MCP Secret.
2

Add a Connector MCP in Twin B

In Twin B, navigate to Connectors and add a new connector:
  • Type: Connector MCP
  • Target URL: https://pria.praxislxp.com/api/mcp
  • Bearer Token: Twin A’s MCP Secret
Connector MCP configuration pointing to Twin A
3

Save and Enable

Save the connector and ensure it is enabled.Connector MCP list showing the enabled connector
4

Test the Connection

Send a query in Twin B that targets Twin A’s knowledge. For example:
Using the pria-mcp, search for information about “onboarding process”
Testing connectivity through the connector MCP
5

Review Tool Details

Check the agent details to confirm the tool was invoked correctly and returned results.Agent details showing MCP tool invocation
The Bearer token you provide in Twin B must match Twin A’s MCP Secret. This ensures authenticated, secure communication between twins.

​
MCP in Real-Time Voice Conversations

MCP tools are available during Convo Mode when using the OpenAI GPT-Realtime voice provider. Your Digital Twin can invoke MCP tools during live voice conversations — for example, searching a connected twin’s knowledge base while speaking with a user.
MCP tool support in Convo Mode is only available with the OpenAI GPT-Realtime provider. ElevenLabs voice sessions do not support MCP tools. See Convo Mode for a provider comparison.

​
Institution-Level MCP Server Configuration

Administrators can configure MCP servers at the institution level through the Configuration page in the Admin dashboard. This makes external MCP tools available to all users of a Digital Twin without requiring individual setup. See Configuration > MCP Servers for details.

​
Troubleshooting

SymptomSolution
401 UnauthorizedVerify your Authorization: Bearer <token> header matches the Digital Twin’s MCP Secret exactly.
Connection refusedConfirm the server URL is https://pria.praxislxp.com/api/mcp and your network allows outbound HTTPS.
Tools not appearingEnsure MCP Server is enabled and saved in the Digital Twin’s Configuration panel.
Twin B ignores the connectorCheck that the server label and tool names are correct. Try an explicit prompt: “Using the pria-mcp, execute search-instance-rag to find information on ‘xyz‘“
Unexpected resultsUse the MCP Inspector to send the same query directly and compare responses.

​
Learn More

  • Configuration — Institution-level MCP server settings
  • Connectors — Managing MCP and other connector types
  • Convo Mode — Voice conversations with MCP tool support
  • Tools — Tool definitions available to your Digital Twin