Open-source LLM knowledge platform: turn raw documents into a queryable RAG, an autonomous reasoning agent, and a self-maintaining Wiki.
Go agentagenticaichatbotdsh-pluginembeddingsevaluationgenerative-aigolangknowledge-basellmmulti-tenantollamaopenaiquestion-answeringragrerankingsemantic-searchvector-searchwiki
import type { ComposerTranslation } from 'vue-i18n'
const discoveryFields = [
'mode', 'servers', 'tools', 'total', 'has_more', 'next_cursor', 'next_step',
'notice', 'status', 'name', 'description', 'input_schema', 'tool_ref', 'server_id',
'server_name',
] as const
function record(value: unknown): Record<string, unknown> {
return value !== null && typeof value === 'object' && !Array.isArray(value)
? value as Record<string, unknown> : {}
}
function discoveryOverlay(data?: unknown): Record<string, unknown> {
const extra = record(data)
const overlay: Record<string, unknown> = {}
for (const key of discoveryFields) {
if (key in extra) overlay[key] = extra[key]
}
return overlay
}
// Discovery historically stored its complete JSON in output, without tool_data.
// Live SSE copies the whole event envelope onto tool_data; only merge catalog fields.
export function parseMcpDiscovery(output?: string, data?: unknown): Record<string, unknown> {
let parsed = {}
try { parsed = record(JSON.parse(output || '')) } catch { /* Error or truncated historical output. */ }
return { ...parsed, ...discoveryOverlay(data) }
}
export function mcpDescriptionLead(text: string): string {
const trimmed = text.trim()
if (!trimmed) return ''
const beforeHeading = trimmed.split(/\n#{1,6}\s+/)[0] || trimmed
const withoutHeadings = beforeHeading.replace(/^#{1,6}\s+/gm, '')
const firstBlock = withoutHeadings.split(/\n\s*\n/)[0] || withoutHeadings
return firstBlock.replace(/\s+/g, ' ').trim()
}
export function mcpDescriptionNeedsExpand(text: string): boolean {
const trimmed = text.trim()
if (!trimmed) return false
if (trimmed.includes('\n')) return true
return [...mcpDescriptionLead(trimmed)].length > 80
}
export function mcpDiscoveryRows(data: Record<string, unknown>) {
const source = data.mode === 'list_servers' ? data.servers : data.tools
if (!Array.isArray(source)) return []
return source.flatMap((value) => {
const row = record(value)
if (typeof row.name !== 'string') return []
return [{
name: row.name,
description: typeof row.usage_instructions === 'string' && row.usage_instructions
? row.usage_instructions : typeof row.description === 'string' ? row.description : '',
status: typeof row.status === 'string' ? row.status : '',
serverName: typeof row.server_name === 'string' ? row.server_name : '',
}]
})
}
export function mcpSchemaParameters(schema: unknown) {
const definition = record(schema)
const required = Array.isArray(definition.required) ? definition.required : []
return Object.entries(record(definition.properties)).map(([name, value]) => {
const property = record(value)
const type = property.type
const variants = property.anyOf || property.oneOf
const types = Array.isArray(variants) ? variants.flatMap((variant) => {
const type = record(variant).type
return typeof type === 'string' ? [type] : []
}) : []
return {
name,
type: typeof type === 'string' ? type : Array.isArray(type) ? type.join(' | ') : types.join(' | '),
required: required.includes(name),
description: typeof property.description === 'string' ? property.description : '',
}
})
}
export function getMcpToolDisplayType(toolName?: string): 'mcp_discovery' | 'mcp_call' | undefined {
if (toolName === 'discover_mcp_tools') return 'mcp_discovery'
if (toolName === 'call_mcp_tool') return 'mcp_call'
return undefined
}
export function mcpToolResultOutput(event: { tool_name?: string; output?: string; error?: string }): string | undefined {
if (!getMcpToolDisplayType(event.tool_name)) return event.output
return event.output || event.error
}
export function getMcpToolTitle(t: ComposerTranslation, event: {
tool_name?: string; arguments?: unknown; output?: string; tool_data?: unknown; pending?: boolean; success?: boolean
}): string {
if (!getMcpToolDisplayType(event.tool_name)) return ''
const data = parseMcpDiscovery(event.output, event.tool_data)
const args = typeof event.arguments === 'string' ? parseMcpDiscovery(event.arguments) : record(event.arguments)
const mode = data.mode || args.mode || ('input_schema' in data ? 'describe' : '')
const keys: Record<string, string> = {
list_servers: 'agentStream.mcp.listServers',
list_tools: 'agentStream.mcp.listTools',
search: 'agentStream.mcp.searchTools',
describe: 'agentStream.mcp.describeTool',
}
const label = t(event.tool_name === 'call_mcp_tool' ? 'agentStream.mcp.callTool' : keys[String(mode)] || 'agentStream.mcp.discoverTools')
const describeName = mode === 'describe' ? data.name || args.tool_name : ''
const serverName = typeof data.server_name === 'string' ? data.server_name : ''
const suffix = describeName || ((mode === 'list_tools' || mode === 'search') ? serverName : '')
const title = suffix && typeof suffix === 'string' ? `${label}:${suffix}` : label
if (event.pending) return t('agentStream.toolStatus.calling', { name: title })
if (event.success === false) return t('agentStream.toolStatus.calledFailed', { name: title })
return title
}