| 1 | use crate::*; |
| 2 | use rmcp::{ |
| 3 | ErrorData, RoleServer, ServerHandler, |
| 4 | model::{ |
| 5 | CallToolRequestParams, CallToolResponse, CallToolResult, ContentBlock, ListToolsResult, |
| 6 | PaginatedRequestParams, ServerCapabilities, ServerConfig, Tool, ToolAnnotations, |
| 7 | }, |
| 8 | service::RequestContext, |
| 9 | }; |
| 10 | |
| 11 | #[derive(Clone)] |
| 12 | struct Observability(Arc<App>); |
| 13 | impl ServerHandler for Observability { |
| 14 | fn get_info(&self) -> ServerConfig { |
| 15 | ServerConfig::new(ServerCapabilities::builder().enable_tools().build()) |
| 16 | } |
| 17 | async fn list_tools( |
| 18 | &self, |
| 19 | _: Option<PaginatedRequestParams>, |
| 20 | _: RequestContext<RoleServer>, |
| 21 | ) -> std::result::Result<ListToolsResult, ErrorData> { |
| 22 | Ok(ListToolsResult { tools: [ |
| 23 | ("get_logs", "Retrieve logs from one granted service."), |
| 24 | ("get_traces", "Find traces from one granted service."), |
| 25 | ("get_trace", "Retrieve a trace with spans limited to granted services."), |
| 26 | ].into_iter().map(|(name, description)| { |
| 27 | let mut properties = json!({"service":{"type":"string"},"q":{"type":"string"},"limit":{"type":"integer","minimum":1,"maximum":500}}); |
| 28 | let required = if name=="get_trace" { properties=json!({"service":{"type":"string"},"trace_id":{"type":"string"}}); json!(["service","trace_id"]) } else {json!(["service"])}; |
| 29 | Tool::new(name, description, json!({"type":"object","properties":properties,"required":required,"additionalProperties":false}).as_object().unwrap().clone()).with_annotations(ToolAnnotations::new().read_only(true)) |
| 30 | }).collect(), ..Default::default() }) |
| 31 | } |
| 32 | async fn call_tool( |
| 33 | &self, |
| 34 | request: CallToolRequestParams, |
| 35 | context: RequestContext<RoleServer>, |
| 36 | ) -> std::result::Result<CallToolResponse, ErrorData> { |
| 37 | let result: Result<Value> = async { |
| 38 | let grant = &context |
| 39 | .extensions |
| 40 | .get::<axum::http::request::Parts>() |
| 41 | .and_then(|p| p.extensions.get::<mcp::Grant>()) |
| 42 | .ok_or_else(|| Error::new(401, "This connection expired. Connect again."))? |
| 43 | .0; |
| 44 | let arguments = request.arguments.unwrap_or_default(); |
| 45 | let service = arguments |
| 46 | .get("service") |
| 47 | .and_then(Value::as_str) |
| 48 | .ok_or_else(|| Error::new(400, "Choose a granted service."))?; |
| 49 | if !array(&grant["resources"]).iter().any(|id| id == service) { |
| 50 | return Err(Error::new( |
| 51 | 403, |
| 52 | "This service is outside the connection's access.", |
| 53 | )); |
| 54 | } |
| 55 | let mut query = HashMap::new(); |
| 56 | for (key, value) in &arguments { |
| 57 | match key.as_str() { |
| 58 | "service" => {} |
| 59 | "q" if request.name != "get_trace" => { |
| 60 | query.insert( |
| 61 | key.clone(), |
| 62 | value |
| 63 | .as_str() |
| 64 | .filter(|s| s.len() <= 4096) |
| 65 | .ok_or_else(|| Error::new(400, "Narrow the search."))? |
| 66 | .to_owned(), |
| 67 | ); |
| 68 | } |
| 69 | "limit" if request.name != "get_trace" => { |
| 70 | query.insert( |
| 71 | key.clone(), |
| 72 | value |
| 73 | .as_u64() |
| 74 | .filter(|n| (1..=500).contains(n)) |
| 75 | .ok_or_else(|| Error::new(400, "Choose a limit from 1 to 500."))? |
| 76 | .to_string(), |
| 77 | ); |
| 78 | } |
| 79 | "trace_id" if request.name == "get_trace" && value.as_str().is_some() => {} |
| 80 | _ => return Err(Error::new(400, "Use the fields listed for this tool.")), |
| 81 | } |
| 82 | } |
| 83 | match request.name.as_ref() { |
| 84 | "get_logs" => { |
| 85 | Ok(json!({"logs":telemetry::logs(self.0.clone(), service, &query).await?})) |
| 86 | } |
| 87 | "get_traces" => { |
| 88 | let jobs = core::scan(self.0.clone()).await?; |
| 89 | let traces = telemetry::traces(self.0.clone(), service, &query, |span| { |
| 90 | visible_span(span, &jobs.value, &grant["resources"]) |
| 91 | }) |
| 92 | .await?; |
| 93 | Ok(json!({"traces":traces})) |
| 94 | } |
| 95 | "get_trace" => { |
| 96 | let id = arguments |
| 97 | .get("trace_id") |
| 98 | .and_then(Value::as_str) |
| 99 | .filter(|s| !s.is_empty() && s.len() <= 128) |
| 100 | .ok_or_else(|| Error::new(400, "Choose a trace ID."))?; |
| 101 | let mut trace = telemetry::trace(self.0.clone(), id).await?; |
| 102 | let jobs = core::scan(self.0.clone()).await?; |
| 103 | if !array(&trace["spans"]) |
| 104 | .iter() |
| 105 | .any(|span| visible_span(span, &jobs.value, &json!([service]))) |
| 106 | { |
| 107 | return Err(Error::new(404, "No trace from that service has this ID.")); |
| 108 | } |
| 109 | trace["spans"] |
| 110 | .as_array_mut() |
| 111 | .unwrap() |
| 112 | .retain(|span| visible_span(span, &jobs.value, &grant["resources"])); |
| 113 | Ok(json!({"trace":trace})) |
| 114 | } |
| 115 | _ => Err(Error::new(404, "No tool with that name.")), |
| 116 | } |
| 117 | } |
| 118 | .await; |
| 119 | Ok(match result { |
| 120 | Ok(value) => CallToolResult::structured(value), |
| 121 | Err(error) => CallToolResult::error(vec![ContentBlock::text(if error.status >= 500 { |
| 122 | "The service couldn't answer. Check its dashboard and retry.".to_owned() |
| 123 | } else { |
| 124 | error.message |
| 125 | })]), |
| 126 | } |
| 127 | .into()) |
| 128 | } |
| 129 | } |
| 130 | fn visible_span(span: &Value, jobs: &Value, resources: &Value) -> bool { |
| 131 | if let Some(id) = span["attributes"]["studio.service"].as_str() { |
| 132 | return array(resources).iter().any(|r| r == id); |
| 133 | } |
| 134 | let owners: Vec<_> = jobs |
| 135 | .as_object() |
| 136 | .into_iter() |
| 137 | .flat_map(|jobs| jobs.iter()) |
| 138 | .filter(|(_, job)| job["job"]["Meta"]["studio_trace_service"] == span["service"]) |
| 139 | .map(|(id, _)| id) |
| 140 | .collect(); |
| 141 | owners.len() == 1 && array(resources).iter().any(|r| r == owners[0]) |
| 142 | } |
| 143 | pub fn router(app: Arc<App>) -> Router { |
| 144 | let state = app.clone(); |
| 145 | mcp::router(app, "observability", move || { |
| 146 | Ok(Observability(state.clone())) |
| 147 | }) |
| 148 | } |
| 149 | |
| 150 | #[cfg(test)] |
| 151 | mod tests { |
| 152 | use super::*; |
| 153 | #[test] |
| 154 | fn trace_grants_reject_foreign_and_ambiguous_spans() { |
| 155 | let jobs = json!({"allowed":{"job":{"Meta":{"studio_trace_service":"backend"}}},"private":{"job":{"Meta":{"studio_trace_service":"private-api"}}}}); |
| 156 | let resources = json!(["allowed"]); |
| 157 | assert!(visible_span( |
| 158 | &json!({"service":"backend","attributes":{}}), |
| 159 | &jobs, |
| 160 | &resources |
| 161 | )); |
| 162 | assert!(!visible_span( |
| 163 | &json!({"service":"private-api","attributes":{}}), |
| 164 | &jobs, |
| 165 | &resources |
| 166 | )); |
| 167 | assert!(!visible_span( |
| 168 | &json!({"service":"backend","attributes":{"studio.service":"private"}}), |
| 169 | &jobs, |
| 170 | &resources |
| 171 | )); |
| 172 | assert!(visible_span( |
| 173 | &json!({"service":"allowed","attributes":{"studio.service":"allowed"}}), |
| 174 | &jobs, |
| 175 | &resources |
| 176 | )); |
| 177 | let mut ambiguous = jobs.clone(); |
| 178 | ambiguous["private"]["job"]["Meta"]["studio_trace_service"] = json!("backend"); |
| 179 | assert!(!visible_span( |
| 180 | &json!({"service":"backend","attributes":{}}), |
| 181 | &ambiguous, |
| 182 | &resources |
| 183 | )); |
| 184 | assert!(!visible_span( |
| 185 | &json!({"service":"unknown","attributes":{}}), |
| 186 | &jobs, |
| 187 | &resources |
| 188 | )); |
| 189 | } |
| 190 | } |