Historian MCP Server

What is MCP?#

MCP, the Model Context Protocol, is an open standard that lets an AI model call tools in another product. An MCP client, which may be a desktop assistant, a coding agent or an orchestration framework, connects to an MCP server, asks it which tools it has, and calls them on the model's behalf. The Historian is such a server: a client connected to it can list datasets, find tags and read time-series data, and the model can reason over the answers.

Every Timebase component publishes an MCP server of the same shape. This page covers the Historian's; MCP Servers Across Timebase lists the others, and Atlas MCP Server covers the one component whose MCP server also takes writes.

Where to find it#

The server is built into the Historian and listens on the Historian's own ports:

Your ports differ if you changed the Historian's Kestrel endpoints. See Component Settings.

What it publishes#

The Historian's MCP server publishes its read tools, the shared discovery tools every component carries, and the log reader. It publishes no prompts; the guidance that used to be delivered as prompts is now carried in the tools' own descriptions and in the built-in skills.

Everything on this list only reads. A model connected over MCP can look at your datasets, tags and data and cannot change them.

The write tools stay off MCP by design. The Historian also has tools that create and update datasets, add and update tags, and write values, and they require the administrator role. They are available to the Historian's own AI assistant when an administrator is signed in, and they are never published at /mcp, whoever connects. To change a dataset or write data from outside, use the Historian's interface or its REST API.

Authentication#

The Historian ships with Pulse authentication switched on, and /mcp is protected by the same rule as the rest of its API: every request needs a Pulse bearer token whose audience includes the Historian. A request without one is answered 401.

The token is a component token, the kind a service obtains from Pulse with a client id and secret. Register a client for your AI tool in Pulse with the Historian among its audiences, as Registering Timebase apps as OAuth clients in Pulse describes, request a token, and send it on every call:

Authorization: Bearer <token>

A component token expires after an hour by default and the header does not renew itself, so a client configured with a fixed header suits scripts and sessions of that length. Nothing on the Historian's MCP server needs a person's sign-in, because nothing on it writes.

With Pulse:Auth:Enabled set to false on the Historian, /mcp is open, like the rest of the API. That is for a local trial, never for a deployment.

Connecting a client#

The steps are the same for every client; only the wording of its settings screen changes.

1. Obtain a Pulse bearer token with the Historian in its audience.
2. Open your AI tool's MCP server settings and add a server.
3. Set the transport to Streamable HTTP.
4. Set the URL to https://<your historian>:4512/mcp.
5. Add a request header named Authorization with the value Bearer <token>.
6. Save and reconnect. The client lists the tools above.

Many clients store this as a small configuration file. The shape below is the common convention, and Claude Desktop, Cursor, Windsurf, VS Code and LibreChat all accept it or a close variant:

{
  "mcpServers": {
    "timebase-historian": {
      "type": "http",
      "url": "https://<your historian>:4512/mcp",
      "headers": {
        "Authorization": "Bearer <token>"
      }
    }
  }
}

Claude Code takes the same from the command line:

claude mcp add --transport http timebase-historian https://<your historian>:4512/mcp \
  --header "Authorization: Bearer <token>"

Postman can connect as well, which is a quick way to see the tool list without an AI model in the loop: create a new MCP request, choose HTTP, enter the URL, add the Authorization header, and connect. Postman lists the tools and lets you call one by hand.

Frameworks that act as MCP clients, such as the OpenAI Agents SDK and LangGraph, take the same URL, transport and header in their own configuration.

Troubleshooting#