Investigate the actual error.
Your agent can inspect grouped issues, the latest stored traceback, route and release context, then examine the relevant code and propose a fix with tests.
Explore error investigationDjango monitoring for you and your coding agent
Give your agent the production context you keep collecting yourself.
Stop copying errors and search reports into chat. FetchNode lets your coding agent retrieve the available evidence from the Django website you’re working on, investigate a problem and propose the next change.
Start monitoring freeOne website free · Inspectable setup · You review changes and deployment
Failed requests
Release v1.8.4Checkout errors after a release
23failed requests
8affected users
v1.8.4release
Failed requests rise after the release. Eight users are linked to the example issue.
Inspect the sample incidentNext: your coding agentExample task
“Investigate the checkout errors after v1.8.4. Inspect the traceback, propose a fix and run the relevant tests.”
You review the changes and deployment.Completed runs
Failures beginA job stopped. Know where to look.
3failed runs
12 minsince last run
Unknowncustomer impact
The scheduled job stopped completing. Customer impact still needs investigation.
Explore job monitoringNext: your coding agentExample task
“Inspect the job’s code and available errors. Propose a fix and checks before I deploy it.”
You review the changes and deployment.Google clicks per day
Search ConsoleSee which page lost search traffic.
840Google clicks
−25%vs. previous 28 days
/guides/page to review
This example page received 840 clicks, down from 1,120. Review the queries before drawing conclusions.
How search context fits inNext: your coding agentExample task
“Review the queries and content for /guides/. Propose improvements and explain which evidence supports them.”
You review the changes and deployment.Illustrative order · sample dataExplore a sample project
Work you can hand over
An investigation can start with evidence.
Find a concrete place to look.
Slow-route measurements, query timing and repeated query patterns help your agent choose a performance investigation. A suspected N+1 pattern still needs verification.
Work through a query examplePropose an informed improvement.
Review declining pages, queries and aggregate landing-page outcomes. Your agent can inspect content and propose changes with a plan to assess new measurements.
Connect search to website outcomesConnect after your first real signal
Use the Django repository you already work in.
First install and configure the official fetchnode Django package with that website’s private server key in its deployment environment. Keep fetchnode in INSTALLED_APPS (fetchnode_client remains supported for existing integrations). Run these commands from the configured repository, using its Python environment.
1. Configure your editor.
python manage.py fetchnode_mcp --setup cursorThis writes Cursor’s MCP settings and updates repository guidance. Other setup targets are windsurf, cline and all. Review the generated files.
2. Verify the connection.
python manage.py fetchnode_mcp --statusThe command checks authentication and reports available tools. A successful connection does not prove your website is sending data; verify a real production signal separately.
3. Set the task.
Reload the agent’s MCP connection, ask it to list the website’s open errors or inspect a measured bottleneck, then request a bounded investigation and relevant checks.
Another MCP client?
python manage.py fetchnode_mcp --print-configInspect the printed stdio configuration and add it to your client. The local command reads the configured key at runtime. Keep private keys out of committed configuration.
The standalone monitoring snippet does not install this management command. If that is your current integration, use the package path in the setup guide before configuring MCP. Monitoring works without an agent.
Understand the access you give
One website’s context. Explicit work.
Read production evidence.
Tools cover errors and stored details, request and database bottlenecks, Search Console, landing-page outcomes and saved SEO advice. Error detail can include stack source lines, internal user IDs and captured request context. Your agent provider may process the results.
Keep project access bounded.
MCP requires a usable private server key or a still-valid legacy key. Public browser keys are rejected. A website’s key cannot retrieve another website’s records. Revoke exposed keys in project settings.
Review actions and code.
MCP includes issue actions to resolve, ignore or snooze. Those change FetchNode issue state; they do not repair or deploy code. Your coding agent’s repository and execution permissions determine what it can change.
FetchNode supplies available measurements, not a guarantee that every problem has been captured. It does not run an autonomous maintenance team, apply dependency upgrades or test every checkout flow. Review the capture and access details before connecting an agent.
Keep room for what you want to build
Start by connecting one real website.
Verify its production data, connect your agent when useful, then repeat the same setup for your next website.
Start monitoring free