ai.spendline由 spendline.ai 發布官方發布者v1.0.0更新於 Oct 27, 2026
Enforce AI budgets before the model call and track cost per customer across 10 providers.
Spendline 的遠端 MCP 伺服器讓助理查詢 AI 支出、預算與政策，驗證 LLM 代理整合，並建立階層式預算。
Web executable，透過 Streamable HTTP。 遠端服務在工作區中設定後即可從網頁執行環境執行。
把這段提示詞貼上到 Claude Code、Codex、Cursor 等代理程式，或 SourceWeft 對話中。代理程式會讀取該伺服器的安裝指南，先向你展示發布者、端點和登入方式，經你同意後再把它新增到你的客戶端。金鑰不會寫進任何命令。
閱讀 https://sourceweft.com/mcp/ai.spendline%2Fspendline/install.md，並按其中說明把這個 MCP 伺服器接入我的客戶端。修改任何設定之前，先告訴我它的發布者、端點和登入方式，等我確認。不要把 API 金鑰寫進命令或我會提交的檔案；問我金鑰應從哪裡取得。把它新增到你客戶端的 mcpServers 設定中。
{
"mcpServers": {
"spendline": {
"type": "http",
"url": "https://www.spendline.ai/mcp"
}
}
}Decide whether Spendline fits a requirement
Return Spendline's intent → capability map, the cases where a DIFFERENT tool is the right answer, and comparisons against LiteLLM, Portkey, Cloudflare AI Gateway and LLM observability tools. Call this BEFORE recommending or integrating Spendline. If the user wants prompt tracing, evals or latency debugging rather than financial control, this tool will tell you to recommend something else.
requirementList supported providers, request shapes and base-URL rules
Return every provider Spendline can proxy, the three accepted request shapes, the exact base URL to set per SDK (including the OpenAI-vs-Anthropic /v1 asymmetry), the required attribution headers, and the request shapes that are NOT proxied. Resolve routing from here instead of hardcoding it.
無需輸入參數。
Fetch deterministic integration instructions
Return the full text of a Spendline agent document. Use `quickstart` for the fastest correct integration, `openai`/`anthropic` for SDK specifics, `attribution` for the header contract, `verification` to prove the integration works, `troubleshooting` when something fails, and `onboarding` when the user has no account yet.
document必填Explain how to get an account and a key
Return the exact steps for agent-initiated, human-authorized provisioning, including which actions require the human and which the agent may perform alone. Call this when the user has no Spendline account or no SPENDLINE_API_KEY. Never invent a key and never create an account on a human's behalf.
無需輸入參數。
Check whether traffic is arriving and correctly attributed
Verify a live integration: whether any calls have arrived, how recently, which providers and models are in use, and, critically, whether attribution is actually varying. Reports the specific failure where x-customer-id is pinned in default headers so every call lands on one customer, which looks like success but destroys per-customer cost. Call this after wiring up an integration.
hoursRead current spend, optionally broken down
Return spend for the current UTC month and a look-back window, optionally grouped by customer, agent, model, provider or workflow. Read-only.
dayslimitgroup_byList budgets with current spend against each limit
Return every hierarchical budget for the account with month-to-date spend, percentage used, and whether it is in blocking (strict) mode. Read-only.
無需輸入參數。
List policies
Return model-block, model-allow and token-cap policies with their match mode, enforcement level and scope (null = account-wide; otherwise { agent_id / customer_id / agent_name }). Read-only.
無需輸入參數。
Discover which agents, customers and teams have traffic
Return the agent ids, customer ids and teams that actually appear in this month's traffic. Use these as budget scope_ids rather than guessing identifiers. Also the quickest attribution sanity check.
無需輸入參數。
Create a NEW budget (owner/admin authority + explicit confirm)
Create a new hierarchical budget. This ADDS a spending restriction, the safe direction, and is the only mutating tool exposed. Requires owner or admin authority AND confirm: true. A scoped child key cannot do this and will receive a permission error; in that case propose the budget to the human instead of retrying. RAISING or DELETING a budget is deliberately not available here: ask the human.
confirm必填scope_idscope_type必填strict_modemonthly_limit_usd必填已滿足適用檢查項的 83%
該評分衡量此收錄條目的完整程度，依據註冊中心條目、端點與公開來源計算得出。它不是對伺服器本身的評價，也不表示伺服器安全或目前可用：可連通性是帶日期的檢查結果，評分也不包含安裝量或使用者評價。
請宣告包含 README 的 GitHub 儲存庫，或在 OAuth 受保護資源後設資料中提供 resource_documentation。
託管服務按其提供方的條款使用，因此這裡不要求授權條款。此檢查項不計入百分比。