thousandeyes-mcp

Model Context Protocol (MCP) server for Cisco ThousandEyes v7 — lets AI assistants query tests, agents, alerts, dashboards, widgets, and test results (network, page-load, web-transactions, path-vis).

Community project - NOT affiliated with Cisco/ThousandEyes.
Status: Alpha (MVP) read-only.

Why (business value)

Faster troubleshooting: Ask AI to “show outages in the last hour” or “figure out where the network traffic is getting stuck at” for super fast issues identification.
Consistency over scripts: Standard MCP tools replace one-off curl snippets.
Safer by default: Read-only; token is only read from env.
Composable workflows: Chain tools (tests → dashboard → widget → test results).

Current capabilities

Tool	What it does	Endpoint(s)
`te_list_tests(aid?, name_contains?, test_type?)`	Lists tests (filter by name/type/AG)	`GET /v7/tests`
`te_list_agents(agent_types?, aid?)`	Lists enterprise / enterprise-cluster / cloud agents	`GET /v7/agents`
`te_get_test_results(test_id, test_type, window?/start?/end?/aid?/agent_id?)`	Test results (e.g., `network`, `page-load`, `web-transactions`; not `dns-server`)	`GET /v7/test-results/{testId}/{testType}`
`te_get_path_vis(test_id, window?/start?/end?/aid?/agent_id?/direction?)`	Path visualization data	`GET /v7/test-results/{testId}/path-vis`
`te_list_dashboards(aid?, title_contains?)`	Lists dashboards	`GET /v7/dashboards`
`te_get_dashboard(dashboard_id, aid?)`	Dashboard details incl. widget list	`GET /v7/dashboards/{dashboardId}`
`te_get_dashboard_widget(dashboard_id, widget_id, window?/start?/end?/aid?)`	Widget data for a dashboard	`GET /v7/dashboards/{dashboardId}/widgets/{widgetId}`
`te_get_users()`	Lists users in the ThousandEyes account	`GET /v7/users`
`te_get_account_groups()`	Lists account groups available to the authenticated org	`GET /v7/account-groups`

Requirements

Python 3.12+
ThousandEyes API v7 bearer token in env: TE_TOKEN

Install

python3 -m pip install -r requirements.txt
# If Python is externally managed:
# python3 -m venv .venv && . .venv/bin/activate && pip install -r requirements.txt

Configure (Claude Desktop)

Add to your claude_desktop_config.json (or Dev UI):

{
  "mcpServers": {
    "thousandeyes": {
      "command": "/ABS/PATH/TO/python3",
      "args": ["/ABS/PATH/TO/repo/src/server.py"],
      "env": { "TE_TOKEN": "YOUR_OAUTH_BEARER_TOKEN" }
    }
  }
}

Token is read only from env, never written to disk.

Try it out!

Example 1: In-depth Analysis of Customer Journeys for Optimizing Performance

TE.MCP.vid.2.mp4

Example 2: Visualizing Performance Data

TE.MCP.vid.1.1.mp4

Here are some other prompts to try out

What was the network health of the Patient Portal between 13:00–14:00 CET on 1 Sep 2025?
Which regions/agents showed elevated page-load time for the Patient Portal between 08:00–10:00 UTC today?”
Show uptime and TTFB for the Patient Portal homepage over the last 24 hours, and call out any drops.
Which enterprise agents had >1% packet loss to api.patient-portal.example.com this morning?”
Open the Patient Portal dashboard and list the widgets relevant to availability and the login flow and callout any widgets reporting no data
For test , compare network latency during 10:00–10:30 UTC vs. the prior 30 minutes.”
Show path visualization anomalies for test around 15:30 UTC yesterday.”

Security & privacy

Read-only tools - no writes.
No tokens or org data stored - token only via TE_TOKEN.
Respect org rate limits - backoff on the roadmap.

Roadmap

Adding feature support for alerts, tags, event detection, endpoint agents, etc.
Optional retries/backoff on 429.
Minimal CI and examples catalog.

License & attribution

Apache-2.0
“ThousandEyes” is a trademark of Cisco Systems, Inc. This project is NOT affiliated with Cisco/ThousandEyes.

Maintainers

Aditya Chellam · Kiran Kabdal