Skip to main content
Glama

ExpenseBot

by io.github.TotesMagotes

submit_receipt

Submit a photo or PDF of a receipt for processing. Covers requests phrased as 'log this', 'log this receipt', 'save this receipt', 'expense this', or 'add this to my expenses', including when the user simply shares a photo of a receipt or invoice. The receipt image is validated, uploaded to cloud storage, and processed by AI to extract vendor, amount, date, tax, and category. The expense appears in the user's spreadsheet in about 1-3 minutes, and longer for PDFs or large batches. Handles images and PDFs, mixed together in one batch.

TO SEND FILES (preferred, and required for PDFs): call this tool with filesToUpload listing every file the user gave you. It returns one signed upload URL per file. Upload them ONE AT A TIME with an HTTP PUT, telling the user which file you just finished and how many remain, then call this tool ONCE with uploadRefs for all of them — that processes the whole set as a single batch, like the ExpenseBot web app. Do not call this tool once per file.

Use the photo parameter for one image or PDF attached in ChatGPT. MCP clients that cannot supply file references may use photoBase64 for one small image; use the upload flow for large files or batches.

Optional note and tag values use the same receipt metadata path as ExpenseBot's camera, file uploader, and forwarded-email intake. The note is stored in the Notes column (L); the tag is stored in the Tag column (K). Batch defaults apply to every file, and each uploadRefs item may override either value for that file.

Input Schema

TableJSON Schema
NameRequiredDescriptionDefault
tagNoBatch-level project or client tag applied to every expense row. Individual files may override it. Stored in the Tag column (K) through the same path as the web receipt uploader.
noteNoBatch-level context note applied to every file in the upload. Individual files may override this with their own note. Stored through the existing receipt-note pipeline in the Notes column (L).
photoNoSingle attached receipt image or PDF supplied by ChatGPT. For multiple files or large files, use filesToUpload and uploadRefs instead.
filenameNoOptional filename (e.g., 'lunch_receipt.jpg')
mimeTypeNoMIME type of the file (default: image/jpeg)
uploadRefNoSingle-file shorthand for uploadRefs. Use uploadRefs when there is more than one file.
uploadRefsNoSubmit previously uploaded files as ONE batch. Include every uploadRef from the filesToUpload step. If any file fails validation the whole batch is rejected and nothing is processed.
photoBase64NoLegacy fallback for MCP clients that send one small image as complete base64 data. ChatGPT should use photo or uploadRefs instead.
filesToUploadNoRequest upload URLs for one or more files. Include EVERY file the user provided in a single call. Returns one signedUrl + uploadRef per file; upload each, then call this tool again with uploadRefs.

Schema Changelog

Changes observed during successful MCP inspections. Dates show when Glama detected each change.

  1. Changed1 schema field changed
    • changedOutput schema / (root)
      Previous value: -{
-  "additionalProperties": true,
-  "description": "Standard ExpenseBot tool result envelope. `message` is the human-readable summary the AI cites; `data` is the structured payload (totals, breakdowns, ids, etc.). On failure, `success` is false and `error` carries a code/message/hint triple.",
-  "properties": {
-    "data": {
-      "additionalProperties": true,
-      "description": "Structured payload. Shape varies per tool — common keys: total, breakdown, comparison, sampleMeta, ids, expenseId, reportId, signupUrl, results.",
-      "type": "object"
-    },
-    "error": {
-      "additionalProperties": true,
-      "description": "Present only when success === false.",
-      "properties": {
-        "code": {
-          "type": "string"
-        },
-        "hint": {
-          "type": "string"
-        },
-        "message": {
-          "type": "string"
-        }
-      },
-      "type": "object"
-    },
-    "message": {
-      "description": "Human-readable result text. Always present on success; prefer rendering this verbatim before any further reasoning.",
-      "type": "string"
-    },
-    "sampleMeta": {
-      "additionalProperties": true,
-      "description": "Set when the underlying dataset was truncated. isTruncated=true means the agent saw a sample of `sampleCount` of `totalCount` rows; aggregate totals are still accurate.",
-      "properties": {
-        "isTruncated": {
-          "type": "boolean"
-        },
-        "sampleCount": {
-          "type": "integer"
-        },
-        "totalCount": {
-          "type": "integer"
-        }
-      },
-      "type": "object"
-    },
-    "success": {
-      "description": "False on tool errors; check before reading `data`.",
-      "type": "boolean"
-    }
-  },
-  "type": "object"
-}New value: +null
  2. Changed5 schema fields changed
    • changedInput schema / properties / photo / description
      Previous value: -"Base64-encoded image or PDF data (JPEG, PNG, HEIC, WebP, or PDF). Max 15MB. Only for small images — large payloads get truncated or dropped in transit; use uploadRef instead."New value: +"Single attached receipt image or PDF supplied by ChatGPT. For multiple files or large files, use filesToUpload and uploadRefs instead."
    • addedInput schema / properties / photo / properties
      Added value: +{
+  "download_url": {
+    "description": "Temporary HTTPS URL supplied by ChatGPT for downloading the attachment.",
+    "format": "uri",
+    "type": "string"
+  },
+  "file_id": {
+    "description": "Attachment identifier supplied by ChatGPT.",
+    "type": "string"
+  },
+  "file_name": {
+    "description": "Original attachment filename supplied by ChatGPT.",
+    "type": "string"
+  },
+  "mime_type": {
+    "description": "Attachment MIME type supplied by ChatGPT.",
+    "type": "string"
+  }
+}
    • addedInput schema / properties / photo / required
      Added value: +[
+  "download_url",
+  "file_id"
+]
    • changedInput schema / properties / photo / type
      Previous value: -"string"New value: +"object"
    • addedInput schema / properties / photoBase64
      Added value: +{
+  "description": "Legacy fallback for MCP clients that send one small image as complete base64 data. ChatGPT should use photo or uploadRefs instead.",
+  "type": "string"
+}
  3. First observed

TDQS

A4.8/5.0
Behavior5/5

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

While annotations indicate readOnlyHint=false and destructiveHint=false, the description goes far beyond that by explaining validation, cloud upload, AI extraction, and the fact that the expense appears in the user's spreadsheet in 1-3 minutes. It also discloses batch rejection behavior (if any file fails validation, the whole batch is rejected and nothing is processed). This gives an agent a realistic model of what will happen after invocation and how to set user expectations.

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?

The description is long but well-structured: it begins with the core purpose, then lays out the preferred upload flow, then covers single-photo and fallback paths, and closes with metadata behavior. Some redundancy exists with the already-detailed schema, but the length is justifiable given the complexity of the two phase upload workflow, nine parameters, and the lack of an output schema.

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 a tool without an output schema and with a complex multi-step upload protocol, the description is nearly fully self-contained: it defines the initial filesToUpload call, the upload actions, the final uploadRefs call, batch rejection behavior, and where note/tag values are stored. It also explains which input parameter to choose for each file and client variant, making 'how to invoke this safely and correctly' very clear. No additional external contract instructions seem missing.

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?

Input schema has 100% per-parameter description coverage, and the schema itself is extremely detailed. The description still adds useful operational semantics beyond the schema, such as the one-at-a-time HTTP PUT upload, the instruction to announce file progress, and the requirement to call the tool once with all uploadRefs, which slightly exceeds the schema's own wording. It doesn't reach 5 because many parameter details in the description are effectively restatements of what the schema already documents.

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?

The description opens with a specific verb and resource: 'Submit a photo or PDF of a receipt for processing.' It then lists exact user phrasings ('log this', 'save this', 'expense this') that this tool should handle, which makes it distinguishable from siblings like add_cash_expense or add_income_from_file. It clearly positions itself as the receipt-image/PDF intake tool rather than a manual expense entry or Gmail scanning tool.

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

Usage Guidelines5/5

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

The description gives explicit when-to-use and when-not-to-use instructions: filesToUpload is preferred and required for PDFs, photo is for a single ChatGPT attachment, photoBase64 is only a fallback, and the user is told never to call the tool once per file. It also explains when to call the tool a second time with uploadRefs and what batch semantics apply, so an agent knows exactly how and when to invoke this tool in its workflow.

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

All ExpenseBot toolsInspect
Try in Browser
RayrunBuild, deploy, and operate MCP servers without becoming an infrastructure team.ray.runAd

Glama MCP Gateway

Add one secure layer between your agents and this server.

TDQS

A3.6/5.0
Disambiguation3/5

Most tools are explicitly scoped, but several analytics/retrieval tools overlap in purpose, such as get_spending_summary vs get_deep_analytics vs get_monthly_books_review, and generic search vs search_expenses vs search_knowledge. The detailed descriptions help, but an agent still has to carefully choose between near-equivalent options like correct_expenses vs update_expense and the three add_income variants.

Naming Consistency5/5

Tool names consistently use lower_snake_case with a recognizable verb prefix: get_*, list_*, add_*, create_*, check_*, scan_*, search_*, and whatif_*. Minor exceptions like fetch and search are still terse retrieval verbs rather than a different naming style, so the overall pattern is predictable.

Tool Count1/5

With 59 tools, this exceeds the 50+ threshold for an extreme tool count and creates a heavy selection surface for an agent. Even though ExpenseBot covers many subdomains, many get_/list_/add_ variants could be consolidated into fewer parameterized tools. The count undermines the otherwise clear naming structure.

Completeness3/5

The surface is strong for creating, reading, and updating expenses, reports, invoices, and Gmail scans, but there are notable lifecycle gaps: no delete/void tools for expenses, income, reports, or invoices, and no update tool for income. Several descriptions explicitly redirect unsupported edits to the web app, confirming that the assistant cannot complete those workflows directly.

Resources