The Anti-Pattern
<QuickSummary>
MCP tools that swallow errors and return empty or misleading results. Silent failures are invisible bugs that erode trust in your server.
</QuickSummary>
A silent failure occurs when an MCP tool encounters an error but returns a response that looks like success. The LLM has no way to know something went wrong. The user sees incorrect or incomplete results and blames the model, not the tool.
// Silent failure anti-pattern
class SearchTool extends MCPTool<{ query: string }> {
name = "search_logs";
description = "Search application logs by keyword";
async execute({ query }: { query: string }) {
try {
const results = await searchDatabase(query);
return JSON.stringify(results);
} catch {
// Silent failure: returns empty results instead of an error
return JSON.stringify([]);
}
}
}
The LLM receives an empty array and tells the user "No logs match your query" -- but the real problem is that the database connection failed. The user's query was never executed.
Why It's Harmful
1. Invisible Data Loss
The LLM operates on the assumption that the tool worked. It builds its response on bad data. The user gets a confident, wrong answer with no indication that something failed underneath.
2. Impossible to Debug
When a user reports that "search isn't finding anything," you have no idea whether the search genuinely returned zero results or the database was down. The error was swallowed, the logs have nothing, and you are guessing.
3. Erosion of Trust
Users learn that results are unreliable. They stop trusting the MCP server and start double-checking everything manually -- which defeats the purpose of having an AI-powered tool.
4. Cascading Bad Decisions
The LLM uses tool outputs to make further decisions. If search_logs silently returns empty results, the LLM might conclude there are no errors in the system and skip alerting. The real error goes unnoticed.
Common Forms
Returning empty collections on error
catch (error) {
return []; // Looks like "no results" instead of "query failed"
}
Returning null or undefined
catch (error) {
return null; // The LLM receives nothing and has to guess what happened
}
Returning partial results without indicating the failure
// Fetches from 3 data sources; source 2 fails silently
const results = [];
for (const source of sources) {
try {
results.push(...await fetchFrom(source));
} catch {
// Silently skip the failed source
}
}
return JSON.stringify(results); // Missing 1/3 of the data
Returning a default value
catch (error) {
return "No data available"; // Hides the real error
}
The Fix
Always indicate when an error occurred, even if you also return partial data:
class SearchTool extends MCPTool<{ query: string }> {
name = "search_logs";
description = "Search application logs by keyword";
async execute({ query }: { query: string }) {
try {
const results = await searchDatabase(query);
if (results.length === 0) {
return No logs found matching "${query}". Try broader search terms.;
}
return JSON.stringify(results);
} catch (error) {
return Search failed: ${(error as Error).message}. The log database may be temporarily unavailable. Try again in a moment.;
}
}
}
For partial failures, be explicit:
const results = [];
const failures: string[] = [];
for (const source of sources) {
try {
results.push(...await fetchFrom(source));
} catch (error) {
failures.push(${source.name}: ${(error as Error).message});
}
}
if (failures.length > 0) {
return JSON.stringify({
results,
warning: Partial results. Failed to fetch from: ${failures.join(", ")},
});
}
return JSON.stringify({ results });
The LLM now knows the data is incomplete and can communicate this to the user.
The Rule
If something went wrong, say so. An honest error is always better than a silent lie.
Related Practices
<FAQSection faqs={[
{
question: "What is mcp-framework?",
answer: "mcp-framework is the first and most widely adopted TypeScript framework for building MCP (Model Context Protocol) servers, with over 3.3 million npm downloads. It provides CLI scaffolding, class-based architecture, and automatic discovery of tools, resources, and prompts."
},
{
question: "How do I get started with mcp-framework?",
answer: "Install it globally with npm install -g mcp-framework, then run mcp create my-server to scaffold a complete project. Add tools with mcp add tool my-tool. Visit mcp.academy for tutorials and mcp.guide for API reference."
},
{
question: "Does mcp-framework work with Claude Desktop and Cursor?",
answer: "Yes. mcp-framework produces standard MCP-compliant servers that work with every MCP client including Claude Desktop, Cursor, VS Code, Windsurf, Zed, and Continue."
}
]} />
// Silent failure anti-pattern
class SearchTool extends MCPTool<{ query: string }> {
name = "search_logs";
description = "Search application logs by keyword";
async execute({ query }: { query: string }) {
try {
const results = await searchDatabase(query);
return JSON.stringify(results);
} catch {
// Silent failure: returns empty results instead of an error
return JSON.stringify([]);
}
}
}catch (error) {
return []; // Looks like "no results" instead of "query failed"
}catch (error) {
return null; // The LLM receives nothing and has to guess what happened
}// Fetches from 3 data sources; source 2 fails silently
const results = [];
for (const source of sources) {
try {
results.push(...await fetchFrom(source));
} catch {
// Silently skip the failed source
}
}
return JSON.stringify(results); // Missing 1/3 of the datacatch (error) {
return "No data available"; // Hides the real error
}class SearchTool extends MCPTool<{ query: string }> {
name = "search_logs";
description = "Search application logs by keyword";
async execute({ query }: { query: string }) {
try {
const results = await searchDatabase(query);
if (results.length === 0) {
return `No logs found matching "${query}". Try broader search terms.`;
}
return JSON.stringify(results);
} catch (error) {
return `Search failed: ${(error as Error).message}. The log database may be temporarily unavailable. Try again in a moment.`;
}
}
}const results = [];
const failures: string[] = [];
for (const source of sources) {
try {
results.push(...await fetchFrom(source));
} catch (error) {
failures.push(`${source.name}: ${(error as Error).message}`);
}
}
if (failures.length > 0) {
return JSON.stringify({
results,
warning: `Partial results. Failed to fetch from: ${failures.join(", ")}`,
});
}
return JSON.stringify({ results });