ai-netcafe

Tables and ledgers checked by arithmetic, not by a model.

CommunityOnlineNo sign-inGlobalFreeRead-only

What it can do

    What data it sees

    Do you need an account

    No: the server works without sign-in

    Tables and ledgers checked by arithmetic, not by a model. 24 tools. MCP 2026-07-28 ready.

    How to connect

    How to connect ai-netcafe to Claude

    The server address to paste into Claude settings:

    https://ainetcafe.com/mcp?s=registry

    1. Open Claude (claude.ai or the desktop app).
    2. Go to Settings → Connectors.
    3. Click Add custom connector.
    4. Paste the server address copied below into Remote MCP server URL and click Add.
    5. In a chat, click + → Connectors and switch the new connector on.

    Custom connectors are available on Free, Pro, Max, Team and Enterprise plans (Free is limited to one). On Team and Enterprise an organization Owner adds the connector first under Organization settings → Connectors.

    Authorization

    After you click Add, a sign-in window for the service opens. Sign in with your own account and approve access. Claude never sees your password.

    How to connect ai-netcafe to ChatGPT

    The server address to paste into ChatGPT settings:

    https://ainetcafe.com/mcp?s=registry

    1. Open ChatGPT in a browser (chatgpt.com). A Plus, Pro, Business, Enterprise or Edu plan is required.
    2. Turn on developer mode once: Settings → Apps → Advanced settings → Developer mode.
    3. Open Settings → Connectors and click Create.
    4. Fill in the form: Name (anything), Description (one line about what the service does), MCP server URL (copy it below).
    5. Under Authentication choose OAuth if the service requires sign-in, otherwise None. Click Create.
    6. In a new chat open + → Apps/Connectors and enable the connector.

    OpenAI has renamed this section before (Connectors → Apps/Plugins). If the label differs, search settings for "developer mode". On Business/Enterprise workspaces an admin must allow custom connectors first.

    Authorization

    On first use ChatGPT opens the service's sign-in window. Sign in and approve access.

    How to connect ai-netcafe to Cursor

    The server address to paste into Cursor settings:

    https://ainetcafe.com/mcp?s=registry

    Fastest: click Open in Cursor below and confirm the prompt.

    Manually:

    1. In Cursor open Settings → Cursor Settings → MCP and click Add new global MCP server.
    2. Paste the JSON copied below into ~/.cursor/mcp.json (per project: .cursor/mcp.json in the repo root).
    3. Save the file. The server appears in the MCP list; authorize it there if asked.

    Authorization

    If the service needs sign-in, an authorize button appears next to the server in the MCP list.

    Server tool list (34)

    Raw names from tools/list. Only developers need these.

    what_can_you_doDescribe a task in plain language (any language) and get back exactly which tools on this server do it, with ready-to-run example calls — instead of reading the whole catalogue and guessing. Also returns multi-step recipes when a task needs several tools chained (invoices to a ledger, a bank statement reconciled, a messy CSV turned into a deliverable). Deterministic and free: it calls no model, costs nothing, and never runs out of quota. Call this FIRST when you are not sure what this server offers.
    list_appsList the open-source AI applications hosted and ready to run at AI NetCafé (ainetcafe.com). Each one normally requires local setup (Docker/Python + your own model API key); here they run pre-configured. Use this to find a tool for a task like translating a PDF with formulas intact, generating a PowerPoint file, polishing an academic paper, or running an autonomous research report. Do not call this first when the request already clearly matches compare_models, translate_pdf, deep_research, or make_slides; call that task tool directly. Example — GET https://ainetcafe.com/t/list_apps
    get_appFull details of one hosted application: what it does, how to use it, measured benchmark scores, source repository, and the URL a human can open to run it. Example — GET https://ainetcafe.com/t/get_app?slug=<slug-from-list_apps>
    ask_modelSend a prompt to one specific large language model and get the answer plus measured platform cost metadata. The beta platform covers the user charge ($0.00); capacity limits still apply. Example — GET https://ainetcafe.com/t/ask_model?prompt=Say+hi&model=deepseek-v4-flash
    compare_modelsRun one prompt across multiple LLMs in parallel and return every answer side by side with measured platform cost metadata and latency. The beta platform covers the user charge ($0.00). This answers "which model should I actually use for this kind of task?" with data instead of guesswork. Example — GET https://ainetcafe.com/t/compare_models?prompt=Explain+CAP+theorem+in+1+line
    list_modelsList every model currently available in the free beta with reference input/output rates and health metadata. Those rates are platform cost metadata only; every user charge is $0.00 during the beta. Example — GET https://ainetcafe.com/t/list_models
    rememberPersist a durable memory: an architecture decision, a stable user preference, a verified bug fix, or an important discovery. The free beta provides a bounded per-caller/workspace memory pool; no personal API key is required. Do not store secrets or raw logs. Example — tools/call remember {"content":"Deploy key rotates monthly"}
    recallRetrieve previously stored memories, optionally filtered by search query and/or project. Call at the start of work on a known project to restore context: why decisions were made, known fixes, preferences. Example — GET https://ainetcafe.com/t/recall?query=<what+to+remember> (needs a workspace/key for durable memory)
    web_searchSearch the live web through a self-hosted SearXNG meta-search (aggregates dozens of engines, no tracking). Returns titles, URLs and snippets. Use when you need current information or sources. Example — GET https://ainetcafe.com/t/web_search?query=latest+MCP+spec
    fetch_pageFetch a public URL and return clean LLM-ready Markdown from the server-rendered response. This tool does not execute browser JavaScript; for SPA or empty-text pages, use web_search, a browser, or the site's API. Use it after web_search to read a reachable public source, or to ingest a static page for analysis. Example — GET https://ainetcafe.com/t/fetch_page?url=https://example.com
    model_costsMeasured platform cost metadata for one call on each model; your charge is $0.00 during the free beta. Vendors publish per-million-token list prices, but a call's cost depends on how many tokens the model chooses to emit — models differ by an order of magnitude on the same prompt. standard_bench sends an IDENTICAL prompt to every model, so the difference is the model, not the workload — use that to choose a model before bulk work. production_mixed is real traffic and is NOT comparable across models. Free to cite, CC BY 4.0. Example — GET https://ainetcafe.com/t/model_costs
    ai_visibilityAudit a URL for AI visibility: which AI crawlers robots.txt actually allows (parsed per user-agent group, not keyword-matched), whether llms.txt / sitemap / JSON-LD / canonical exist, and how much real text an agent gets without running JavaScript. Returns a score plus the specific fixes, ordered by impact.
    pdf_to_markdownConvert a PDF (or a scanned page image) into clean Markdown that keeps headings, lists and tables, and puts multi-column pages in the right reading order. Text-layer PDFs are read exactly and cost far less; images go through a vision model.
    extract_tablesExtract tables from a PDF into structured rows (JSON + CSV). Pass fields to force a fixed set of columns — that aligns a pile of documents that each name their headers differently into one consistent table. Rows the model was unsure about are flagged rather than guessed. Text-layer PDFs only.
    extract_statementTurn a bank statement or transaction PDF into a clean transaction table (JSON + CSV), then cross-check it: opening + credits - debits must equal the stated closing balance. If it does not balance you get the exact difference and which row the running balance first breaks at — so you know whether the table is safe to use for accounting. Text-layer PDFs only (scanned images not yet supported).
    json_yamlConverts JSON to YAML or YAML to JSON. It works out which one you gave it, so you do not have to say. A parse failure comes back with the parser message instead of silently producing something that looks fine and is not. Use when a config, a CI file, or a Kubernetes manifest needs to be in the other format.
    validate_jsonChecks that text parses as JSON, and optionally that required keys are present with the right top-level types. Returns the specific violations, not just true/false. Checks required + types only — not full JSON Schema, and it says so rather than pretending. Use before feeding generated JSON into something that will fail on it.
    diff_textReturns which lines were added and which were removed, with line numbers — computed with a longest-common-subsequence, not guessed by a model. Use to compare two versions of a config, a document, or any command output, instead of asking an LLM to eyeball two blobs and hoping it notices.
    jwt_decodeDecodes the header and payload of a JWT and reports issued-at / expiry as readable timestamps plus seconds remaining. The signature is NOT verified and the response says so — decoding is fine for debugging a token you already hold, but never treat these values as proof of anything; verification needs the secret and belongs in your own service.
    regex_testRuns a regular expression against sample text and returns every match with its position and capture groups (named groups included). Use before wiring a pattern into code, instead of guessing whether the escaping survived the trip through JSON and the shell.
    diff_tablesMatches rows across two CSVs on a key column and reports three things: keys only in A, keys only in B, and keys in both whose other columns disagree — naming the exact column and both values. Unlike reconcile_ledger this needs no amount column, so it also fits name lists, inventory counts, permission tables, and any "these two exports should match" check.
    clean_tableTidies a spreadsheet export: removes duplicate rows, trims whitespace (half-width and full-width — Chinese exports are full of 　), unifies the half-dozen ways a cell can say "empty" (NA / null / - / 无), drops empty rows and columns, and can split one column into several. Returns the cleaned CSV plus exactly what changed: rows in, rows out, duplicates removed, cells trimmed per column. It can also transpose rows/columns and unpivot a wide table into a long one. The row arithmetic is verified in code — if in − removed ≠ out, the response says so instead of handing back a table nobody can check. Use when a CSV came out of Excel or an export and needs cleaning before analysis.
    merge_tablesCombines up to 20 CSVs into a single table. Headers do not have to match: columns are unioned and a file missing a column contributes blanks for it, so rows never shift silently — the failure mode that makes hand-merged spreadsheets untrustworthy. Reports each source file row count and checks in code that they sum to the merged total. Use for monthly exports, per-store sheets, or any set of files with the same subject but drifting headers.
    reconcile_ledgerReconciles two sets of records — your books against a bank, platform, or supplier statement. Matches rows on a key column, compares an amount column, and returns three lists: only in A, only in B, and same key but different amount. Amounts are compared in integer cents, so 0.1 + 0.2 never invents a phantom difference for someone to chase. The response also proves the result: the listed differences are re-added and must equal the gap between the two totals, checked in code. Use for month-end close, platform payouts vs orders, or any "these two numbers should agree and do not" problem. This is the job people do by hand with VLOOKUP or a groupby and then cannot prove they got right.
    extract_invoicesGive it up to 20 invoice URLs (PDF or page images) and get back one table ready to post: number, date, seller, buyer, net / tax / gross, currency. Every row is checked in code — net + tax must equal gross — and the batch total is re-added independently, so a row the model misread is flagged with the exact difference instead of quietly landing in your books. Mixed currencies get no batch total on purpose: adding them together would be an accounting error. CSV is UTF-8 with BOM so Excel opens it right.
    create_taskCreate a task that runs on a schedule in our cloud — you do not keep anything running. It only notifies you when the result actually changes. Kinds: watch_page (Watch a web page and report when its content changes); daily_answer (Re-run a web-researched question on a schedule and report when the answer changes); watch_reachability (Track whether a site stays reachable from mainland China); pipeline (Run one of your production lines (create_pipeline) on a schedule; every run leaves a proof-carrying work order). Needs a workspace token (?w=ws_... on your MCP URL) so you can manage it later. Application and model calls are subsidized during the free beta; your charge is $0.00 and capacity limits apply.
    list_tasksShow scheduled tasks, next run times, run counts, and measured platform cost metadata. User charge is $0.00 during the beta.
    get_task_runsRecent runs of one scheduled task: what it returned, whether the result changed, and measured platform cost metadata. User charge is $0.00.
    delete_taskStop and remove a scheduled task and its run history.
    transpile_sqlConvert a SQL statement from one dialect to another — mysql, postgres, sqlite, tsql, oracle, snowflake, bigquery, redshift, spark, hive, presto, trino, duckdb, clickhouse, databricks, doris, starrocks and more. Deterministic parser (sqlglot), not an LLM: the same input always produces the same output, and syntax errors come back with the exact line and column. Use it when migrating queries between databases or debugging dialect-specific syntax.
    china_reachabilityFetch a URL from a real mainland-China network egress and report HTTP status, latency and China DNS resolution. Answers "is my site/API usable from China?" with a measurement instead of a guess — you cannot get this from a VPS abroad.
    render_diagramTurn diagram-as-code into an image: Mermaid, PlantUML, Graphviz/DOT, C4, Excalidraw and 20+ more (self-hosted Kroki). Returns a hosted SVG/PNG URL you can embed directly in Markdown or HTML. Example — GET "https://ainetcafe.com/t/render_diagram?source=graph TD;A--%3EB&format=png"
    check_jobGet the status or result of a job started by deep_research, translate_pdf, or make_slides. Poll every 15-30 seconds until status is "done" or "error". While work is pending, follow retry_after_seconds and next_action; when complete, prefer structured_result when present. Example — GET https://ainetcafe.com/t/check_job?job_id=<id-from-a-job-tool>
    build_appTurn one plain-language description into a LIVE single-page web tool: code is generated, deployed to managed hosting with HTTPS, and listed — you get the public URL in ~1-2 minutes. Best for tool-style apps: calculators, converters, checklists, timers, generators, small games. Async — poll with check_job. Example — tools/call build_app {"description":"a tip calculator web app"} → poll check_job