Connect your AI agent to the live story map.

Use the official server endpoint below when adding StoriesOnBoard as a remote MCP connector. The general availability announcement lists Standard, Pro, Plus, Premium, and trial plans. Check the current plan information and your agent host's connector support before setup.

For Claude Desktop, the documented path is Settings → Connectors → Add custom connector. Enter the endpoint, connect, choose the StoriesOnBoard workspace, and authorize it. Configure the tools you want the host to make available, then start a conversation. Other hosts may use a different setup screen; use their supported remote MCP connection flow rather than assuming the same configuration file.

The server exposes discovery, read, and write operations. Start by finding the intended map and inspecting its current structure. Do not infer identifiers, valid field values, or positions from a screenshot when they can be read from the map.

The examples below describe tasks rather than promise a fixed inventory. Consult the current tool reference and the tools exposed in your connected host for accepted inputs and available operations.

Begin with a small task that proves you connected the correct workspace and map. Ask the agent to list maps, identify the one you mean, and summarize a selected journey without writing anything. Review the map name and the resulting summary before moving to edits.

For our meal-ordering example, the agent should recognize that Confirm address sits in the payment step of the first release. If it describes a different board or cannot read the story details, resolve the connection or access issue before asking it to generate new work.

Use a bounded write task next. Ask for a proposed description of one existing story, review the text, and explicitly authorize the change you want. The agent should then inspect the updated card rather than treating a submitted request as proof of the final state.

Human approval here is a workflow instruction, not a claim that every host always prompts before every write. Review your host's tool permissions. If an operation fails or the connection drops, read the current state before retrying; an uncertain result is a reason to inspect, not blindly repeat a creation request.

If no maps appear, verify the authorized workspace and account. If reads work but a write fails, inspect the error and the tool configuration instead of repeatedly rewriting the prompt. A visible title alone is not proof of write access.

When the agent identifies the wrong card, narrow the task with the map and surrounding step. When results are incomplete, ask what was inspected and which details were unavailable. The connection supplies a way to interact with the product; the host, model, instructions, permissions, and source material still affect the outcome.