Use AI agents with Spotflow
Connect your AI agent to Spotflow with the MCP server to integrate firmware, investigate crashes, and work with your device fleet.
Bring device logs, metrics, and crash data into the AI agent you already use, such as Claude, ChatGPT, Codex, or Cursor. Spotflow provides a remote Model Context Protocol (MCP) server, so your agent can integrate your devices, investigate issues with real field data, and work with your fleet without leaving your editor or terminal.
Set Up Spotflow with One Prompt
Copy this prompt into a coding agent that can fetch web pages and connect to remote MCP servers:
Set up Spotflow for me. Fetch https://spotflow.io/agent-setup/prompt.md and follow the instructions appropriate for my project.The setup instructions guide the agent through connecting the MCP server, integrating your firmware using the documentation for your platform, and verifying that telemetry arrives. The agent follows only the steps relevant to your project; you authorize the connection and approve the changes.
If you prefer to set up the connection yourself, continue with the steps below.
Connect Your Agent Manually
Add the MCP Server
Add the Spotflow MCP server to your agent's MCP settings:
https://api.spotflow.io/mcpThe server uses the Streamable HTTP transport (not the legacy SSE transport). Check your client's documentation for how to add a remote MCP server.
Every request to the server requires authentication, including connecting and listing tools. Sign in with OAuth if your client supports it, or use an API key otherwise.
Authenticate
Sign In with OAuth
Recommended for interactive use. In a client that supports remote MCP OAuth, connect to the server URL and follow the browser sign-in flow. Sign in with your Spotflow account, then select the workspaces and permissions your agent may use. You do not need to paste any key into the conversation.
Connect with an API Key
For automation and clients without OAuth support, create a user or workspace API key with only the scopes your agent needs. A workspace API key can access only its workspace; a user API key can access the workspaces its owner is a member of.
Configure your client to send the key with every MCP request:
Authorization: Bearer <SPOTFLOW_API_KEY>Store the key using your client's secret storage or environment variables. Never paste it into a prompt or commit it to your repository.
Device ingest keys cannot authenticate the MCP connection. Devices connect with ingest keys, while tools and people use API keys or OAuth.
Try It Out
Ask your agent which Spotflow workspaces it can access, or try one of the example prompts below.
What Your Agent Can Do
Once connected, your agent can work with most of what is available in the Spotflow web application and the Management API:
- Query telemetry: search logs, crash reports, and system events, and query built-in and custom metrics.
- Inspect devices: list and filter devices in your fleet.
- Investigate crashes: read the captured fault, stack frames, threads, registers, and variables from core dump analysis.
- Search the documentation: find integration instructions for your platform in this documentation.
- Manage your fleet: work with dashboards, alert rules, firmware versions, symbol files, SBOMs and vulnerabilities, OTA deployments, and ingest keys.
To keep your agent's context small, the server exposes a few frequently used tools directly. The agent discovers the remaining tools by searching for them when it needs them.
Every tool call is checked against the permissions you granted. Read-only access is enough to investigate issues; changes such as creating dashboards or deployments require the corresponding write permissions.
Example Prompts
What your agent can find depends on your integration, the telemetry your devices send, and the permissions you granted. Always review the agent's conclusions and proposed changes.
Integrate Your Firmware
Let your coding agent find the right SDK instructions for your platform, update your project, and help verify that telemetry arrives.
Integrate Spotflow into this firmware project. Identify the platform, follow the relevant docs, and edit the project to collect logs and built-in metrics. Identify useful custom metrics for this application and instrument them too. Build and test the changes, and explain how to verify that telemetry reaches Spotflow.Investigate a Crash
Connect field logs to captured stack frames, registers, and variables. Give your agent the evidence it needs alongside your source code.
Investigate the latest crash in my Spotflow workspace. Inspect the logs and available core dump analysis, relate the evidence to this codebase, and suggest a fix.Find Firmware Regressions
Query logs and metrics across your fleet to compare firmware versions and identify where reliability changed.
Compare crashes and available reliability metrics for the two latest firmware versions in my Spotflow workspace over the last seven days. Identify regressions and affected devices, and show the supporting queries and any gaps in the data.Summarize Fleet Health
Get an engineering summary of reliability trends, recurring errors, and the next steps that matter most.
Summarize the last seven days in my Spotflow workspace. Highlight reliability trends, recurring errors, affected devices, and changes worth investigating. Give me a short engineering summary with the most important next steps.Build Dashboards
Describe what matters to your product. Your agent can discover metrics and prepare a dashboard with your approval.
Find the metrics available in my Spotflow workspace and propose a fleet-health dashboard.Triage Vulnerabilities
Explore SBOMs and vulnerability findings to understand which firmware versions and devices need attention.
Review the vulnerability findings in my Spotflow workspace. Identify affected firmware versions and devices, summarize the evidence, and propose a triage plan without changing any findings.Keep Your Agent Safe
- Grant only what the task needs. Select only the relevant workspaces and permissions when signing in, or create an API key with only the required scopes. Read-only access is enough for investigations.
- Approve changes. Review actions that modify your workspace, such as deployments or alert rules, before your agent runs them. Do not let it flash hardware or start deployments without your approval.
- Keep credentials out of the conversation. Never paste API keys or ingest keys into prompts. When your agent creates an ingest key, the key is not shown in the conversation; the agent receives a short-lived link to save it to a file instead.