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
/**
* Pure helpers for the "WeKnora as MCP server" integration page. Kept free
* of Vue so the snippet builders can be unit-tested with node:test.
*/
export const MCP_TOKEN_PLACEHOLDER = '<YOUR_TOKEN>'
/**
* Derive the public MCP URL from the displayed API base (".../api/v1") and
* the endpoint path ("/mcp/<id>") returned by the backend.
*/
export function buildMcpEndpointUrl(apiBaseUrl: string, endpointPath: string): string {
const base = (apiBaseUrl || '').trim().replace(/\/+$/, '').replace(/\/api\/v1$/, '')
const path = endpointPath.startsWith('/') ? endpointPath : `/${endpointPath}`
return `${base}${path}`
}
/**
* Sanitize an endpoint name into an mcpServers key. Names that leave no
* ASCII characters (e.g. Chinese) fall back to the endpoint id so two such
* endpoints never collide in the client configuration.
*/
export function mcpServerKey(name: string, id = ''): string {
const slug = (name || '')
.trim()
.toLowerCase()
.replace(/[^a-z0-9]+/g, '-')
.replace(/^-+|-+$/g, '')
if (slug) return `weknora-${slug}`
const idSlug = (id || '').replace(/[^a-z0-9]/gi, '').slice(0, 8).toLowerCase()
return idSlug ? `weknora-${idSlug}` : 'weknora'
}
/** Cursor-compatible JSON for a native Streamable HTTP connection. Other clients use their own setup format. */
export function buildHttpClientSnippet(name: string, url: string, token: string, id = ''): string {
return JSON.stringify(
{
mcpServers: {
[mcpServerKey(name, id)]: {
url,
headers: { Authorization: `Bearer ${token || MCP_TOKEN_PLACEHOLDER}` },
},
},
},
null,
2,
)
}
/** Ready-to-copy workspace .vscode/mcp.json for a native HTTP connection. */
export function buildVsCodeClientSnippet(name: string, url: string, token: string, id = ''): string {
return JSON.stringify(
{
servers: {
[mcpServerKey(name, id)]: {
type: 'http',
url,
headers: { Authorization: `Bearer ${token || MCP_TOKEN_PLACEHOLDER}` },
},
},
},
null,
2,
)
}
/** One-liner for Claude Code. */
export function buildClaudeCodeCommand(name: string, url: string, token: string, id = ''): string {
const auth = `Authorization: Bearer ${token || MCP_TOKEN_PLACEHOLDER}`
return `claude mcp add --transport http ${mcpServerKey(name, id)} ${url} --header "${auth}"`
}
/** Fallback when native HTTP/Bearer setup is unavailable, including Claude Desktop without Request headers. */
export function buildStdioBridgeSnippet(name: string, url: string, token: string, id = ''): string {
return JSON.stringify(
{
mcpServers: {
[mcpServerKey(name, id)]: {
command: 'npx',
args: ['-y', '[email protected]', url, '--header', 'Authorization:${WEKNORA_MCP_AUTH}'],
env: { WEKNORA_MCP_AUTH: `Bearer ${token || MCP_TOKEN_PLACEHOLDER}` },
},
},
},
null,
2,
)
}
export interface ToolCatalogEntry {
name: string
group: string
destructive: boolean
}
export interface GroupedTools<T extends ToolCatalogEntry> {
group: string
tools: T[]
}
/** Group catalog entries in the backend's group order, dropping empty groups. */
export function groupTools<T extends ToolCatalogEntry>(groups: string[], tools: T[]): GroupedTools<T>[] {
const known = new Set(groups)
const order = [...groups, ...tools.map((t) => t.group).filter((g) => !known.has(g))]
const out: GroupedTools<T>[] = []
for (const group of Array.from(new Set(order))) {
const members = tools.filter((t) => t.group === group)
if (members.length > 0) out.push({ group, tools: members })
}
return out
}
/** Summarize a knowledge-base scope for the card: empty means "all". */
export function summarizeScope(ids: string[] | null | undefined, names: Record<string, string>): string[] {
return (ids || []).map((id) => names[id] || id)
}