Skip to main content
Glama

Check the numbers in a text against the file

verify_numbers

Check every number in a piece of text (a report, an email, a summary, or an answer you wrote) against the spreadsheet it came from. Each number is counted again from the file and comes back as matches, counted_differently, differs (with the value the file gives) or not_checked, each with how it was counted. Use it before you send or show numbers taken from a file. Fixed rules do the reading and counting; no AI reads the text or the file unless you set ai_recheck, and no AI ever sees the file's rows. Send text, and the file as file_url (a Google Sheet shared as 'Anyone with the link', a Dropbox link, or a file in a public GitHub repository) or as file_content (CSV text, or base64 for Excel) with file_name. No account needed. The file is not stored. The text and the result are kept for 24 hours at a link that anyone with the link can open. Limits: text up to 20,000 characters and 200 numbers, file up to 10 MB, 10 checks per 10 minutes and 40 per day from one place.

Tool call template

Replace each placeholder with a value matching the input schema. Only required parameters are included; add optional parameters as needed.

{
  "id": 1,
  "jsonrpc": "2.0",
  "method": "tools/call",
  "params": {
    "arguments": {
      "text": "<text>"
    },
    "name": "verify_numbers"
  }
}

Input Schema

NameRequiredDescriptionDefault
fileNoA file the person attached in ChatGPT. ChatGPT fills this in. Other assistants send file_url or file_content instead.
textYesThe text whose numbers should be checked, as it will be sent or shown: a report, an email, a summary, or an answer you wrote. Up to 20,000 characters and 200 numbers. Keep the sentences around the numbers; they say what each number counts.
sheetNoFor an Excel file with several sheets, the name of the sheet to read. Default: the sheet with the most data.
file_urlNoA link to the file the numbers came from: a Google Sheet shared as 'Anyone with the link', a Dropbox link, a file in a public GitHub repository or gist, or an Every Number sample such as https://everynumber.app/samples/tickets/tickets-export-2026-10-18.csv. Send this or file_content.
file_nameNoThe file's name with its extension, such as orders-october.csv or pipeline.xlsx. Needed with file_content.
ai_recheckNoOff unless set to true. Then each number that comes back as differs (up to 8) is read once more: a small Claude model gets its sentence with that number hidden, the text's first two sentences, and the file's column names and date range, never the file or its rows, and answers with how to count it. Our code counts that. The number changes to matches only if the count is exactly the number written; otherwise it stays differs. Adds a few seconds.
file_contentNoThe file itself, up to 10 MB: CSV or TSV text as it is, or an Excel file encoded as base64. Send this or file_url.
content_encodingNoHow file_content is written. Default: base64 for .xlsx, .xlsm and .xls, text for everything else.

Output Schema

NameRequiredDescriptionDefault
fileYes
numbersYesEvery number found in the text, in the order written.
summaryYes
accuracyYesHow well this check did the last time we measured it on reports it had never seen.
ai_recheckYesFilled only when you set ai_recheck.
expires_atYesWhen the text and the result are deleted (ISO 8601). The file itself is never stored.
result_urlYesThe same result as a page: the text with every number marked, and how each was counted. Anyone with this link can open it until it's deleted.
text_scopeYesThe basis the text as a whole counts on. Worked out only when several of its numbers agree with it.
how_checkedYes

Schema Changelog

Changes observed during successful MCP inspections.

  1. Changed5 schema fields changed
    • addedInput schema / properties / ai_recheck
      Added value: +{
      +  "description": "Off unless set to true. Then each number that comes back as differs (up to 8) is read once more: a small Claude model gets its sentence with that number hidden, the text's first two sentences, and the file's column names and date range, never the file or its rows, and answers with how to count it. Our code counts that. The number changes to matches only if the count is exactly the number written; otherwise it stays differs. Adds a few seconds.",
      +  "type": "boolean"
      +}
    • addedOutput schema / properties / ai_recheck
      Added value: +{
      +  "anyOf": [
      +    {
      +      "additionalProperties": false,
      +      "properties": {
      +        "asked": {
      +          "description": "How many differs numbers the AI model read again.",
      +          "maximum": 9007199254740991,
      +          "minimum": -9007199254740991,
      +          "type": "integer"
      +        },
      +        "now_match": {
      +          "description": "How many of them now match.",
      +          "maximum": 9007199254740991,
      +          "minimum": -9007199254740991,
      +          "type": "integer"
      +        },
      +        "skipped": {
      +          "anyOf": [
      +            {
      +              "type": "string"
      +            },
      +            {
      +              "type": "null"
      +            }
      +          ],
      +          "description": "Why the read didn't run, when it didn't."
      +        }
      +      },
      +      "required": [
      +        "asked",
      +        "now_match",
      +        "skipped"
      +      ],
      +      "type": "object"
      +    },
      +    {
      +      "type": "null"
      +    }
      +  ],
      +  "description": "Filled only when you set ai_recheck."
      +}
    • addedOutput schema / properties / numbers / items / properties / read_by_ai
      Added value: +{
      +  "description": "True when a small AI model read this sentence (ai_recheck) and our code's count of that reading is the written number. False for everything fixed rules read alone.",
      +  "type": "boolean"
      +}
    • changedOutput schema / properties / numbers / items / required
      Previous value: -[
      -  "text",
      -  "at",
      -  "context",
      -  "verdict",
      -  "how",
      -  "file_says",
      -  "note"
      -]New value: +[
      +  "text",
      +  "at",
      +  "context",
      +  "verdict",
      +  "how",
      +  "file_says",
      +  "note",
      +  "read_by_ai"
      +]
    • changedOutput schema / required
      Previous value: -[
      -  "result_url",
      -  "expires_at",
      -  "summary",
      -  "file",
      -  "text_scope",
      -  "numbers",
      -  "how_checked",
      -  "accuracy"
      -]New value: +[
      +  "result_url",
      +  "expires_at",
      +  "summary",
      +  "file",
      +  "text_scope",
      +  "numbers",
      +  "ai_recheck",
      +  "how_checked",
      +  "accuracy"
      +]
  2. Changed2 schema fields changed
    • addedOutput schema / properties / file / properties / column_names_not_read
      Added value: +{
      +  "description": "Column names the check couldn't read; it reads English ones. Filled only when most can't be read, and then no number is called different: numbers that don't match are not_checked.",
      +  "items": {
      +    "type": "string"
      +  },
      +  "type": "array"
      +}
    • changedOutput schema / properties / file / required
      Previous value: -[
      -  "name",
      -  "rows_read",
      -  "sha256",
      -  "rows_left_out"
      -]New value: +[
      +  "name",
      +  "rows_read",
      +  "sha256",
      +  "rows_left_out",
      +  "column_names_not_read"
      +]
  3. Changed1 schema field changed
    • changedOutput schema / properties / numbers / items / properties / note / description
      Previous value: -"For differs: another way of counting that does give the written number, when there is exactly one."New value: +"For differs: another way of counting that does give the written number, when there is exactly one. For counted_differently: set when the written number only comes out with one more condition than the sentence states (how has that condition, file_says has the value as the sentence reads)."
  4. Added

TDQS

A4.6/5.0
Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Goes well beyond the annotations: it discloses that fixed rules (not AI) do the counting, that no AI sees the file's rows, that the file is not stored, that text and results persist for 24 hours at a publicly open link, that no account is needed, and that ai_recheck is off by default with its exact behavior described. It also quantifies limits (20,000 chars, 200 numbers, 10 MB, 10 per 10 minutes, 40 per day), which an agent cannot infer from readOnlyHint/destructiveHint=false alone.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

Front-loaded with the core action and result categories, then usage, then inputs, then privacy and limits — a logical order. It is a dense single block with long compound sentences (the ai_recheck and link-retention clauses in particular), so a little structural trimming would help, but essentially every sentence carries information.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

For an 8-parameter, nested-object tool with an output schema, the description covers what the schema cannot: privacy posture, retention window, rate and size limits, and the deterministic-vs-AI counting distinction. Return values are summarized as outcome categories even though the output schema exists, leaving no material gap for an agent deciding to call it.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema coverage is already 100%, so the baseline is 3, but the description adds routing meaning: file_url is for Google Sheets/Dropbox/public GitHub links, file_content is CSV text or base64 for Excel and requires file_name, and ai_recheck's escalation path is explained. It stops short of explaining sheet or content_encoding defaults, which remain schema-only.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

States a specific verb and resource: 'Check every number in a piece of text ... against the spreadsheet it came from', and names the artifact types (report, email, summary, answer). It also lists the possible outcomes (matches, counted_differently, differs, not_checked), which pins down the operation precisely enough to distinguish it from a dashboard-building sibling.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines4/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

Gives a clear triggering condition: 'Use it before you send or show numbers taken from a file.' However, it never names an alternative tool or an exclusion case, and does not say when verification is unnecessary (e.g., numbers not originating from a file).

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

Try in Browser

Glama MCP Gateway

Add one secure layer between your agents and this server.

Resources