NextTrace MCP
Use this skill when the user wants route diagnostics through NextTrace or asks how an Agent should call NextTrace.
Start MCP
NextTrace exposes MCP only as a deploy submode over HTTP. There is no stdio MCP mode.
nexttrace --deploy --mcp
nexttrace --deploy --mcp --listen 0.0.0.0:1080 --deploy-token "$TOKEN"
Endpoint:
http://127.0.0.1:1080/mcp
Auth:
- Loopback listen addresses are tokenless by default unless a token is set.
- External listen addresses require a token. If none is provided, NextTrace prints a generated token to stdout.
- MCP/API clients authenticate with
Authorization: Bearer <token> or X-NextTrace-Token: <token>.
- Browser WebUI uses
/auth/login and an HttpOnly cookie.
- Do not put tokens in URL query strings.
Workflow
- Call
nexttrace_capabilities first when available.
- Pick the narrowest common tool for the job:
- Local route:
nexttrace_traceroute
- Repeated loss/latency stats:
nexttrace_mtr_report
- Probe-level stream records:
nexttrace_mtr_raw
- Path MTU:
nexttrace_mtu_trace
- Global vantage points:
nexttrace_globalping_trace
- Other tools:
nexttrace_speed_test, nexttrace_annotate_ips, nexttrace_geo_lookup, nexttrace_globalping_limits, nexttrace_globalping_get_measurement
- Prefer
structuredContent; use text content only as a fallback.
- Preserve explicit user inputs:
target, protocol, port, source_address, source_device, ASN, location, and ip_version. Do not substitute them unless the user asks for a fallback.
- On errors or missing results, report the exact failure and suggested next step. Do not automatically switch protocol, port, location, ASN, tool, or local/Globalping mode.
- For full tool schemas and boundaries, read references/mcp-tools.md and references/capability-matrix.md.
- For Globalping, read references/globalping.md. For local source/device/TOS behavior, read references/platform-notes.md. Use references/cli-fallback.md only when MCP is unavailable, unsupported, or the user asks for CLI output.
- Before writing the final answer, use references/output-templates.md for a concise Markdown shape.
- Keep this skill and its references synced with
server/mcp.go whenever MCP tools or parameters change.
References
- MCP tools: tool names, inputs, outputs, unsupported parameter families.
- Globalping: worldwide traceroute workflow and comparison guidance.
- Capability matrix: local vs Globalping vs CLI fallback.
- Output templates: bilingual Markdown report shapes for each MCP tool.
- CLI fallback: commands to use when MCP is unavailable.
- Platform notes: OS-specific source, device, TOS, raw-socket details.
- Validation: smoke checks and expected behavior.
1---2name: nexttrace3description: Use NextTrace through its deploy MCP endpoint for traceroute, MTR, MTU discovery, speed tests, IP annotation, GeoIP lookup, and Globalping multi-location traceroute. Trigger when an agent needs network path diagnostics or needs to call NextTrace MCP tools.4---56# NextTrace MCP78Use this skill when the user wants route diagnostics through NextTrace or asks how an Agent should call NextTrace.910## Start MCP1112NextTrace exposes MCP only as a deploy submode over HTTP. There is no stdio MCP mode.1314```bash15nexttrace --deploy --mcp16nexttrace --deploy --mcp --listen 0.0.0.0:1080 --deploy-token "$TOKEN"17```1819Endpoint:2021```text22http://127.0.0.1:1080/mcp23```2425Auth:2627- Loopback listen addresses are tokenless by default unless a token is set.28- External listen addresses require a token. If none is provided, NextTrace prints a generated token to stdout.29- MCP/API clients authenticate with `Authorization: Bearer <token>` or `X-NextTrace-Token: <token>`.30- Browser WebUI uses `/auth/login` and an HttpOnly cookie.31- Do not put tokens in URL query strings.3233## Workflow34351. Call `nexttrace_capabilities` first when available.362. Pick the narrowest common tool for the job:37 - Local route: `nexttrace_traceroute`38 - Repeated loss/latency stats: `nexttrace_mtr_report`39 - Probe-level stream records: `nexttrace_mtr_raw`40 - Path MTU: `nexttrace_mtu_trace`41 - Global vantage points: `nexttrace_globalping_trace`42 - Other tools: `nexttrace_speed_test`, `nexttrace_annotate_ips`, `nexttrace_geo_lookup`, `nexttrace_globalping_limits`, `nexttrace_globalping_get_measurement`433. Prefer `structuredContent`; use text content only as a fallback.444. Preserve explicit user inputs: `target`, `protocol`, `port`, `source_address`, `source_device`, ASN, location, and `ip_version`. Do not substitute them unless the user asks for a fallback.455. On errors or missing results, report the exact failure and suggested next step. Do not automatically switch protocol, port, location, ASN, tool, or local/Globalping mode.466. For full tool schemas and boundaries, read [references/mcp-tools.md](references/mcp-tools.md) and [references/capability-matrix.md](references/capability-matrix.md).477. For Globalping, read [references/globalping.md](references/globalping.md). For local source/device/TOS behavior, read [references/platform-notes.md](references/platform-notes.md). Use [references/cli-fallback.md](references/cli-fallback.md) only when MCP is unavailable, unsupported, or the user asks for CLI output.488. Before writing the final answer, use [references/output-templates.md](references/output-templates.md) for a concise Markdown shape.499. Keep this skill and its references synced with `server/mcp.go` whenever MCP tools or parameters change.5051## References5253- [MCP tools](references/mcp-tools.md): tool names, inputs, outputs, unsupported parameter families.54- [Globalping](references/globalping.md): worldwide traceroute workflow and comparison guidance.55- [Capability matrix](references/capability-matrix.md): local vs Globalping vs CLI fallback.56- [Output templates](references/output-templates.md): bilingual Markdown report shapes for each MCP tool.57- [CLI fallback](references/cli-fallback.md): commands to use when MCP is unavailable.58- [Platform notes](references/platform-notes.md): OS-specific source, device, TOS, raw-socket details.59- [Validation](references/validation.md): smoke checks and expected behavior.