Connectors

Connectors let agents read patient data from outside Medbot. A connector is an MCP server, such as a FHIR gateway, a document store or a lab system. Once connected, Medbot lists its tools, and agents that use the connector can call the tools you allow while they run.

Go to Datasets → Connectors (medbot.me/datasets/connectors ).

Kinds of connector

Adding a connector

1. Click Add connector.

2. Enter a Name (for example Northside FHIR).

3. Choose What it serves: FHIR server, Unstructured notes or Other MCP server.

4. Enter the MCP server URL (http or https), for example https://fhir.example.org/mcp.

5. Choose the Authentication:

   • Bearer token: enter the Token.
   • Header: enter the Secret and the Header name (default X-Api-Key).
   • None.

   The secret is stored encrypted and is never shown again.

6. Organisation admins can tick Add it for <organisation> (ticked by default for them): every member’s agents can use it, and admins manage it. Otherwise the connector is personal.

7. Click Connect. Medbot creates the connector and tests the connection straight away.

The connector card

Each card shows the connector’s kind, whether it belongs to the organisation or is personal, and its status: Connected, Not reachable (with the last error) or Not tested yet. It lists the tools agents may call, and how many agents use it and when it was last checked.

• Test connection connects to the server again, lists its tools and updates the status.
• Settings (managers only) opens the connector’s settings.
• Extraction (notes connectors) opens the extraction run.

Settings and allowed tools

In a connector’s settings you can change its Name and MCP server URL, Replace the secret (leave it blank to keep the current one), and choose the Tools agents may call. Tools that look like they change data (create, update, delete, send and so on) are tagged action and are unticked by default.

Until you save the settings once, agents may call every tool the server lists, including action tools. Open Settings, choose the tools, and click Save to restrict them. Test the connection first if no tools are listed.

Changing the URL or secret resets the status to Not tested yet. Remove connector deletes it; agents using it lose its tools.

Who can manage connectors

Using a connector in an agent

Choose connectors in step 2 of the agent wizard, up to five per agent. On every run, each connector’s allowed tools are added to the agent’s tools, named after the connector (for example northside_fhir__search), so renaming a connector changes its tools’ names. Results from a tool are capped at 64,000 characters. A connector that has never been tested, or that the person running the agent may not use, adds no tools.

Notes extraction

A notes connector can turn a store of free-text notes into structured records in a dataset of its own, so agents and validation can use them.

Setting it up

In the notes connector’s Settings, under Extraction:

1. Choose the Tool that lists notes. It should return notes with an id and a patient key.
2. Optionally choose the Tool that reads one note. It is called with { id } when a note’s text is not in the list.
3. Tick the categories to extract:

Running it

Select the connector’s Extraction panel and click Run extraction (managers only). The panel shows progress (done of total patients), counts by category, and a log. Every extracted item keeps the note it came from, so a clinician can check it.

• A run reads the notes, then extracts up to 200 patients, 10 at a time.
• Pause extraction stops the run after the current batch of 10.
• Run extraction again starts a fresh run from the first patient. Re-extraction replaces each patient’s previous AI-derived items.
• Only one run per connector can go at a time.

The results go into a dataset named <connector> (extracted), with source Notes connector, owned by the manager who first ran extraction. Click Review extracted patients to open it. When a run finishes, you get an extraction finished notification.