Roboto MCP Server#

Overview#

The Roboto MCP server at mcp.roboto.ai brings Roboto’s AI tools to external AI applications: Claude Code, Claude Desktop, Codex, Cursor, VS Code, and any other client that speaks the Model Context Protocol. Connect it once, and the AI you already work in can search your Roboto data, explore datasets and files, and analyze topic data mid-conversation, using the same capabilities that power Roboto’s own agents.

The server is the outbound counterpart of MCP Connections: MCP Connections bring external tools into Roboto’s AI; the Roboto MCP server brings Roboto’s tools to external AI apps.

Key properties:

  • You authenticate as yourself. Each connection is tied to your Roboto account, and every tool call runs with your own permissions, never anyone else’s.

  • Read-only, apart from tools that create something new. Nothing exposed over MCP modifies or deletes data you already have. The tools that write only ever add a record, as you: a visualizer layout or workspace, so the AI can hand you a link that opens your data with panels laid out, or a copy of a thread you can carry on separately. Sending a thread a message goes further — Roboto’s AI answers as you, doing anything it can do in AI Chat. Every tool that writes advertises itself as one to your client.

  • All of your organizations, one connection. If you belong to multiple organizations, a single connection reaches them all; you choose which one is active (see Working across organizations).

For setup instructions per client, see the Use the Roboto MCP Server user guide.

Authentication#

The server supports two ways to authenticate:

  • Sign in with your browser (OAuth). The default for interactive clients. When you add the server, your client opens Roboto’s sign-in page in a browser; approving the sign-in connects the client. A sign-in remains valid for up to 30 days, after which the client prompts you to sign in again.

  • Personal access token. For non-interactive use (scripts, CI, or headless agents), send a Roboto access token in the Authorization header instead. The token must have full API access: scope-restricted tokens are rejected, because MCP tools work across the same API surface the token restricts.

Either way, the identity is the same: a token and a browser sign-in for the same account see the same data.

Working across organizations#

A connection maps to your user account, not to one organization. Each tool call operates in your active organization:

  • If you belong to one organization, it is always active and needs no configuration.

  • If you belong to several, ask the AI to call whoami to see your organizations and which one is active, and set_active_org to switch. Roboto never guesses: until you choose, org-scoped tools return an error asking you to pick. Your choice persists for 30 days from when you set it, across conversations and reconnects.

Two overrides pin an organization without touching the active one: every org-scoped tool accepts an optional org_id argument for a single call, and clients that can send custom headers may set X-Roboto-Org-Id on the connection to fix the organization for every call. When both are present, the per-call org_id wins; either override wins over the active organization.

What the AI can do#

Connected over MCP, the AI works with the same capabilities that power Roboto’s own agents:

  • Search your data — find datasets, files, topics, and events with plain-English queries, exact-term lookups, or generated RoboQL, informed by the tags and metadata keys in use across your organization.

  • Look up any record — datasets, files, topics, events, sessions, devices, actions, triggers, invocations, collections, custom fields, and metric definitions, along with AI-generated summaries and comments.

  • Follow the relationships between them — a dataset’s files, topics, events, sessions, and invocations; a file’s topics and events; a device’s sessions; an action’s invocations; a collection’s members. The same tool takes stock of an organization as a whole: its collections, devices, actions, triggers, custom fields, and metric definitions.

  • Trace a collection’s history — how its membership, tags, and custom-field values changed from one version to the next.

  • Read file content — inspect text files like READMEs, configs, CSVs, and logs.

  • Analyze topic data — statistics, anomalies, and patterns in time-series signals, plus links that open the underlying data in the Roboto visualizer when pasted into a comment or AI Chat.

  • Analyze media — scout keyframes across a video topic, then ask a vision model about specific time ranges.

  • Open data in the visualizer — save a visualizer workspace (the files or datasets to open plus the panels to show) or a reusable layout, and get a link that opens it in the Roboto web app. Saving creates only the record it saves; nothing else is touched.

  • Work with threads — read a Roboto thread, fork it into a copy of your own to carry on separately, or send it a message and wait for the reply. Sending a message runs Roboto’s AI as you, so it can do anything it can do in AI Chat — which, unlike the rest of this surface, includes writing.

  • Convert and explain — convert timestamps between epoch and calendar formats, and answer questions about Roboto itself from its documentation.

Two session tools exist only on the MCP surface: whoami reports who you’re signed in as, your organizations, and which one is active, and set_active_org switches the active organization (see Working across organizations).

Every tool is self-describing. Your MCP client fetches the current tool list with full parameter documentation when it connects, so the client itself always shows the up-to-date surface — or ask the AI what Roboto tools it has.

The analysis tools accept optional scope_start_time and scope_end_time arguments (nanoseconds since the Unix epoch) that restrict the analysis to a time window, so you can focus on one maneuver or incident inside a long recording.

Security and permissions#

  • Permission-scoped: tools see exactly what your Roboto account can see, enforced by the platform, not by the AI client.

  • Organization boundary: the active organization is always one you’re a member of, and tools return data only from it.

  • Read-only, apart from tools that create something new: no tool exposed over MCP modifies or deletes data you already have. The ones that write — saving a visualizer layout or workspace, forking a thread — add only that new record, as you. Sending a thread a message is the one call that reaches further: it runs Roboto’s AI on your behalf, which can do anything you can do in AI Chat. Every tool that writes advertises itself to your client as a writing tool, so your client can tell you which is which before you approve a call.

  • Revocable: deleting a personal access token cuts off that token’s access immediately. A browser sign-in expires on its own after at most 30 days, and you can disconnect the server from your client at any time.

On this page