CheckMCP
‹ directory
57MCP SCORE
D

Desearch

https://mcp.desearch.ai/mcp
39 tools~20k tok/req100% annot.proto 2024-11-05
View public page ›
Xin
Raw JSON ›

Score by pillar

Security ·2075
Tool design ·1835
Schemas / desc ·1680
Reliability ·14100
Context-cost ·1217
Compliance ·1260
Coverage ·870

reliability: LOW (T1 single-shot, not credited - requires T3 >=24h)

Context-cost

20ktokens for tools/list

Paid on every request. Lean footprint for 39 tools.

this server20k
median≈ 3k

MCP primitive coverage

the quality score covers tools
39
Tools
0
Resources · none
0
Prompts · none

Why this score — causal attribution

Each penalty: measure → mechanism → effect → Δscore.

Context-cost · HIGH−10.6

~19k tokens (≥p95) → tool defs are paid on EVERY request → exhausts the window + ongoing cost

Security · CRITICAL−5

[MCP06] lethal trifecta: untrusted-content ingestion + sensitive-data access + exfil/destruction -> an injection can exfiltrate → MCP security flaw → tool: (serveur)

Compliance · MEDIUM−1.7

JSON-RPC-compliant errors 1/2 → errors not compliant with the JSON-RPC spec → clients break unpredictably

Compliance · MEDIUM−1.6

proto 2024-11-05 (4 rev. behind 2025-11-25) → outdated protocol version → recent features unavailable, recent clients weakened

Compliance · MEDIUM−1.2

capability 'resources' declared but resources/list empty → capabilities inconsistent with real behavior → clients misled

Compliance · MEDIUM−1.2

capability 'prompts' declared but prompts/list empty → inconsistent capabilities → clients misled

Tool design · MEDIUM−0.5

1 clusters consolidables → list/get/search/... repeated on the same object → candidates to merge into composite tools

Security — OWASP MCP Top 10

MCP06CRITICAL(serveur)

Suggested composite optimizations

ai-search, x-search, web-search, search → search (outil unique avec param `action`, ou sous-serveur namespacé)
x-search ↔ search, web-links-search ↔ x-links-search, x-posts-by-urls ↔ x-posts-by-user, people-search-status ↔ people-search-pause → renommer pour distinguer

Badge — paste it in your README

MCP Score badge — Desearch[![MCP Score](https://checkmcp.dev/badge/mcp-desearch-ai.svg?url=https%3A%2F%2Fmcp.desearch.ai%2Fmcp)](https://checkmcp.dev/report?url=https%3A%2F%2Fmcp.desearch.ai%2Fmcp)
Own this server?Get verified — fresh re-audit, badge if you pass ›

About Desearch — FAQ

Is the Desearch MCP server safe to use?+
CheckMCP audited Desearch and gave it an MCP Score of 57/100 (grade D) — weak — significant issues; review the findings before trusting it with sensitive data or tools. The audit runs an OWASP MCP Top 10 security pass (tool poisoning, hardcoded secrets, command injection, the lethal trifecta) against the live endpoint; see the per-pillar breakdown and the "why this score" attribution on this page, and re-audit anytime at checkmcp.dev.
What is the MCP Score of Desearch?+
Desearch scores 57/100 (grade D) on CheckMCP's vendor-neutral audit across six weighted pillars — security, tool design, schemas, context-cost, compliance and coverage — with reliability measured and shown but not yet credited.
How many tools does Desearch expose, and what is its context cost?+
Desearch exposes 39 tools, and its tools/list response costs roughly 20k tokens — paid on every request your agent makes to it. Lower is better for your context budget.
How was Desearch scored?+
By probing the live MCP endpoint (https://mcp.desearch.ai/mcp), inspecting its tools, schemas and protocol compliance, running an OWASP MCP Top 10 security pass, and measuring the token cost of tools/list — then attributing every penalty as measure → mechanism → effect. The methodology is open: checkmcp.dev/#methodology.