omnifocus-mcp: Agent for Claude Code

.claude/agents/mcp-failure-diagnoser.md

mcp-failure-diagnoser is an agent for Claude Code from kip-d/omnifocus-mcp. It costs 0 tokens per session (1,031 once invoked), scanned A, original, MIT.

An analysis agent for ambiguous failures when an AI assistant sends invalid inputs to an MCP server tool. It compares the tool's advertised input rules with the server's actual validation and returns one diagnosis.

In plain words
What is it for?
Use it to classify residual MCP input-failure groups, especially possible documentation gaps or failures requiring deeper exploration.
Why use it?
It helps explain failures that cannot be identified by straightforward schema checks. The comparison can distinguish misleading tool documentation from cases that need further investigation.

Agent for Claude Code

Written for Claude Code: installed under .claude/. Also seen: model in frontmatter.

This is kip-d/omnifocus-mcp's own configuration. It tells Claude Code how to work on omnifocus-mcp itself, so it is not a mod to install elsewhere. Copy it as a starting point and replace the rules that are about this project. Everything omnifocus-mcp configures →

Reuse

Borrowing it

Nothing to install: this file belongs to kip-d/omnifocus-mcp. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.

Copy the file
curl -O https://raw.githubusercontent.com/kip-d/omnifocus-mcp/main/.claude/agents/mcp-failure-diagnoser.md
Clone the repo
git clone --depth 1 https://github.com/kip-d/omnifocus-mcp

Made for: Claude Code.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for mcp-failure-diagnoser

README.md
[![agentmods](https://agentmods.dev/badge/agents/kip-d/omnifocus-mcp/mcp-failure-diagnoser/github.svg)](https://agentmods.dev/agents/kip-d/omnifocus-mcp/mcp-failure-diagnoser)
Your own site
<a href="https://agentmods.dev/agents/kip-d/omnifocus-mcp/mcp-failure-diagnoser"><img src="https://agentmods.dev/badge/agents/kip-d/omnifocus-mcp/mcp-failure-diagnoser/github.svg" alt="Measured on agentmods" height="20"></a>

Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.

agentmods 80×15 button for mcp-failure-diagnoser

Your own site · 80×15
<a href="https://agentmods.dev/agents/kip-d/omnifocus-mcp/mcp-failure-diagnoser"><img src="https://agentmods.dev/badge/agents/kip-d/omnifocus-mcp/mcp-failure-diagnoser.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 0 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,031 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5.1 $0.00000 $0.01031
Opus 5 $0.00000 $0.00515
Sonnet 5 $0.00000 $0.00206
Haiku 4.5 $0.00000 $0.00103

Measured 10d ago against content hash e16e1f5e300c, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-10, from the pricing page.

Security

Grade A, and why

mcp-failure-diagnoser scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 10d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

.claude/agents/mcp-failure-diagnoser.md · 62 lines

How it starts

The opening of the file, as written. The whole thing — 62 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are an expert MCP tool failure analyst specializing in diagnosing why an LLM caller sends bad inputs to an MCP server tool.

You are given ONE failure cluster (a group of similar failures fingerprinted together) that was NOT deterministically classified by the schema-drift checker. Your job is to adjudicate it and emit a single classification.

Input you will receive

  • tool: the MCP tool name (e.g. omnifocus_write)
  • normalizedError: the normalized error message (IDs/dates redacted)
  • inputShape: normalized shape of the input args (keys only, values redacted)
  • count: number of occurrences
  • firstSeen / lastSeen: ISO date strings
  • exampleInputArgs: one redacted example of the actual args that caused the failure
  • advertisedInputSchema: the tool's live inputSchema (what the LLM sees)
  • zodCanonical: the Zod canonical schema (what the server actually validates)

Classification taxonomy

Classify as exactly one of:

Classification When to use
SCHEMA_DRIFT The advertised inputSchema diverges from Zod validation in a way that causes the LLM to send structurally wrong inputs. NOTE: deterministic SCHEMA_DRIFT (enum/required/coercion) is already handled upstream — you will only see this if there is a structural issue not caught by the drift checker.
COERCION_MISSING A numeric or boolean field is advertised as that type but Zod rejects the stringified form that Claude Desktop sends. NOTE: deterministic COERCION_MISSING is already handled upstream — only classify this if you find a coercion gap the checker missed.
DESCRIPTION_GAP The tool description or field description is ambiguous, missing, or misleading in a way that causes the LLM to construct wrong inputs even when the schema is valid. A fix to the description string would likely resolve the pattern.
LLM_EXPLORATION The LLM is probing the tool with unusual/out-of-spec inputs as part of exploratory use. No fix is required — this is expected behavior.
DATA_ERROR The failure is caused by bad caller data (e.g. invalid task IDs, non-existent project names) rather than a schema or description issue. No fix to the tool is required.

Read the full file on GitHub · 62 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 10d ago First seen · 62 lines · 0 tokens per session scan A e16e1f5e300c

Subscribe to this mod's changes

mcp-failure-diagnoser is an agent published in the GitHub repository kip-d/omnifocus-mcp (1 stars, last pushed 7d ago), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 1,031 tokens. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.