mcpField
M
MCP Advisor

MCP Advisor

FeaturedLocal
github.com/istarwyhDeveloper Tools89Updated 2d ago

About MCP Advisor

MCP Advisor is an MCP server that helps MCP clients find servers for a task and get installation guidance. It combines search providers to return recommendations, but does not install servers, edit client configuration, or verify server safety.

Key features

  1. 01Recommend MCP servers by task
  2. 02Accept optional keyword and capability filters
  3. 03Aggregate multiple search providers
  4. 04Return installation guidance text
  5. 05Offer optional offline retrieval and reranking

Use cases

  1. 01Find MCP servers for a specific task
  2. 02Compare candidates matching keywords or capabilities
  3. 03Get instructions for manual installation and setup
  4. 04Search a server catalog in offline environments

MCP support

As an MCP server, it provides client tools for server recommendations and installation guidance. The guidance is text only; it does not install servers or modify client configuration.

Configuration

Local server config is the same for every client. Copy it into your client’s MCP config file.

mcp.json · Universal config
{
"mcpServers": {
"mcpadvisor": {
"command": "npx",
"args": [
"-y",
"@xiaohui-wang/mcpadvisor"
]
}
}
}
Run commandnpx -y @xiaohui-wang/mcpadvisor

Local servers need a runtime on your machine (e.g. Node.js 18+). Config file locations differ by client; see your client’s docs.

Packages
npm@xiaohui-wang/mcpadvisor

Tools

2

Saved tool list · updated 3w ago

ParameterTypeRequiredDescription
querystringYesPlease provide a precise description of the MCP server you need. Examples of valid queries: - 'MCP server for deploying risk control strategies' - 'MCP server for actuarial pricing of insurance products' Examples of invalid queries: - 'Insurance MCP server' (too broad) - 'Risk control system' (lacks a specific insurance scenario) - 'Actuarial tool' (does not specify the required functionality) The query should clearly specify: 1. Business process (e.g., product pricing, underwriting, claims, reserve calculation, etc.) 2. Specific functional requirements (e.g., risk analysis, strategy deployment, strategy development, feature development, etc.)

Details

From the project README

MCP Advisor

English | 简体中文

MCP Advisor helps an MCP client discover servers for a task and obtain installation guidance. It aggregates search providers and returns recommendations; it does not install a server, edit your client configuration, or verify a recommended server's safety for you.

Start here

  1. Use the current source: follow the Quick Start for MCP 2026-07-28 clients, stdio, or HTTP.
  2. Need an older client? The published npm version 1.0.5 is a separate legacy release. Do not apply the current source's transport configuration to it.
  3. Develop or diagnose: read Contributing, Architecture, and Technical Reference. For current protocol failures, start with the migration guide.
Award and project background

MCP Advisor is developed by the mcpadvisor team, which won second prize at the inaugural OceanBase AI Hackathon in 2025 (首届 OceanBase AI 黑客松大赛). The organizer's May 19 event report names the team and describes its MCP search and recommendation project; the event concluded in Guangzhou on May 17, 2025. The author's OceanBase community post connects that project to this repository. This is historical recognition, not a benchmark or endorsement of every current feature.

What is available?

CapabilityCurrent-source status
recommend-mcp-serversTask-based recommendations with optional keywords/capabilities
install-mcp-serverInstallation guidance/configuration text only; no automatic installation
Meilisearch, Compass, GetMCPConstructed by the CLI; external availability/configuration still matters
Offline providerEnabled by SearchService by default; embeddings can require a model download
Nacos discoveryOpt-in; CLI requires address, username and password
Cline MarketplaceImplemented, opt-in via CLINE_MARKETPLACE_ENABLED=true
FlashRank rerankingImplemented, default off; separately started Python service and model
Search-quality evaluationEvaluation tooling and illustrative fixtures; not proof of production quality
Automatic installation / quality guaranteesNot implemented / not established

The active Meilisearch provider uses text search. Offline retrieval combines text and vectors; this is not universal vector search, adaptive query weighting, or a guarantee of five results. Online providers still run alongside offline data.

Choose a runtime

Current source (modern protocol)

Use Node.js 20+ for the application; Inspector 2.10.1 additionally requires Node.js 22.19+. Use the pinned pnpm 8.15.9 from package.json.

git clone https://github.com/istarwyh/mcpadvisor.git
cd mcpadvisor
corepack enable
pnpm install --frozen-lockfile
pnpm run build
node build/index.js

The default is stdio. Diagnostics must not go to stdout. For local HTTP:

TRANSPORT_TYPE=http SERVER_HOST=127.0.0.1 SERVER_PORT=3000 node build/index.js

Connect a compatible MCP 2026-07-28 client to http://127.0.0.1:3000/mcp. See the official SDK client example. HTTP is stateless, with a fresh protocol server per request; provider caches can remain process-scoped. Old SSE/REST modes are rejected and old HTTP paths return 410. Modern POST responses may still use SSE encoding. There is no built-in authentication or cross-origin browser CORS: keep loopback binding unless you provide an independently secured gateway. Hosting acceptance is not implied.

Published npm 1.0.5 (legacy)
npx -y @xiaohui-wang/[email protected]

This selects the older published artifact, not the source above. Check its own client compatibility and configuration. A source checkout still declaring version 1.0.5 does not mean those source changes have been published to npm.

Third-party installers

Marketplace/Smithery listings and their installers are independently maintained. Check the selected artifact/version, permissions, and transport before running them. This guide does not certify that they deliver the current modern source.

Configure and extend

The CLI does not read SEARCH_PROVIDER, SEARCH_LIMIT, or SEARCH_MIN_SIMILARITY. Provider injection and SearchOptions belong to the programmatic SearchService API, not invented environment flags. Do not import SearchService from the npm root: the package entry is a CLI with startup side effects. See the source-checkout API examples.

TopicGuide
CLI, HTTP, Nacos credentials and programmatic optionsQuick Start
Local Meilisearch lifecycleLocal Meilisearch
Text-search behaviorFull-text search
Optional Cline sourceCline Marketplace
Optional local model service, failure fallback and privacyFlashRank
Judgments, metrics and illustrative-data limitsSearch-quality evaluation
Inspector prerequisites and focused browser checksInspector UI testing
Modern protocol and rollback boundariesMCP migration
Older operational diagnosticsTroubleshooting

Troubleshooting's hosting-related revision is tracked separately in PR #40; use the migration guide for current transport behavior. No unmerged hosting configuration is part of these instructions.

Validate a change

pnpm run check
pnpm test
pnpm run build
pnpm run test:protocol

These cover lint/format, Vitest, production compilation, and focused offline protocol/startup checks respectively. They do not prove live provider availability or ranking quality. The aggregate Playwright suite is pnpm run test:e2e (scripts/e2e/e2e.run.sh); it needs its documented browser/service prerequisites. A selected Inspector spec is not the aggregate suite. See Contributing and Inspector UI testing.

Project direction and documentation

Roadmap preserves historical plans, not current delivery dates. Cline and FlashRank are implemented optional features, not open implementation TODOs. New provider integrations, quality claims, and deployment work need their own scope and evidence.

The documentation review record identifies the source baseline, verification boundaries, and language-mirror policy. Contributions are welcome through an issue and a focused, tested pull request.

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.