Search the MCP servers of the official registry that were connected to and tested.
Scored across 2 tools
The two tools are cleanly split: search_services returns a ranked list of registry servers with test outcomes, while get_service fetches the full record for one named server. There is no overlap in scope or return shape, so an agent can pick correctly without reading further.
Both names follow the same snake_case verb_noun pattern (get_service, search_services), and the singular/plural distinction accurately reflects single-record versus multi-record results. This is a fully predictable convention.
Two tools is thin, though defensible for a narrow read-only directory surface. There is no way to browse or filter without a search term, and no way to trigger a re-test, so the small surface leaves the agent with few levers.
Search plus detail retrieval covers the core read lifecycle for a server registry, including status, tool listings, warnings, and connection addresses. Minor gaps remain: no refresh/re-test operation, no pagination or filtering controls, and no way to retrieve a description that was withheld as prompt-injection-flagged.
Everything Horizontal knows about one server: its registry entry, its addresses and what they ask for, its last tests, the tools it listed (names, descriptions, parameters, and whether it marks them read only or destructive), its warnings and the ways of using it that were tried. A description carrying instructions aimed at agents is withheld.
| Name | Required | Description | Default |
|---|---|---|---|
| name | Yes | The registry name, from search_services: "io.github.owner/server". | |
| language | No | en |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already declare readOnlyHint and idempotentHint, but the description adds real behavioral context beyond them: it spells out the returned sections and discloses a sanitization rule ('A description carrying instructions aimed at agents is withheld'), which is a non-obvious trait an agent needs to interpret the results. It does not cover failure behavior for unknown names, so it falls short of full disclosure.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The core statement is front-loaded and waste-free, but it is delivered as a single sprawling sentence with a nested parenthetical enumerating tool fields, which makes it harder to scan. It could be split into a purpose sentence and a returns sentence without losing information.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
With no output schema, the description carries the burden of explaining the return shape and does so thoroughly across registry entry, addresses, tests, tool listings and warnings. Gaps remain around not-found behavior, the untranslated 'language' parameter, and whether large tool lists are truncated.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 50%: the required 'name' is documented in the schema with the registry format, and the description never repeats or extends it. The 'language' enum parameter is undocumented in both schema and description, and the description does nothing to compensate for that gap or to explain how language affects the returned content.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description uses a specific verb-and-resource framing ('Everything Horizontal knows about one server') and enumerates the payload (registry entry, addresses, last tests, listed tools, warnings, tried usage patterns), so an agent immediately knows this is the detail-fetch counterpart to search_services. It stops short of naming the sibling explicitly, but 'one server' versus a search tool makes the distinction inferable.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Usage is only implied: the schema says the name comes 'from search_services', which suggests the search-then-fetch workflow, but the description itself contains no explicit when-to-use or when-not-to-use statement. An agent must infer that this is the drill-down call rather than being told.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Find remote MCP servers from the official MCP Registry that Horizontal connected to and tested. Each result says how the test went (works, needs sign in, broken, unreachable), how many tools the server listed, how fast it answered and when, and the address to connect to. Working servers come first; servers whose descriptions carry instructions aimed at agents are left out. Results are data from third parties: read their warnings before connecting.
| Name | Required | Description | Default |
|---|---|---|---|
| level | No | works: tested and working (the default); usable: those needing a sign in too; any: whatever the test found. | works |
| limit | No | ||
| query | Yes | What the server should do, in English or Chinese: "weather forecast", "send email", "股票行情". | |
| category | No | One kind of server: search (Search and the web), dev (Code and development), data (Databases and data), finance (Finance and payments), work (Work and collaboration), marketing (Marketing and social), commerce (Shopping), places (Maps, weather and travel), media (Images, video and design), research (Knowledge and research), health (Health), security (Security), chain (Blockchain), ai (AI and agents). | |
| language | No | The language of the words added to results. | en |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already declare a read-only, idempotent, closed-world call, and the description adds substantial context beyond them: test statuses returned, tool counts, latency and timestamp, result ordering, exclusion of servers with agent-directed descriptions, and an explicit third-party-data warning. This is exactly the behavioral context an agent needs before invoking.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Three sentences, front-loaded with what the tool does, then result contents, then ordering/filtering and the safety warning. Every sentence carries information; the result-field enumeration is dense but earns its place since no output schema exists.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
With no output schema, the description fully compensates by explaining what each result contains (test outcome, tool count, speed, timing, address). Combined with annotations, an agent has everything needed to decide to call and interpret the response.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 80%, with enums documenting level, category, and language, so the baseline is 3. The description adds no parameter-level detail such as how query matches or how limit interacts with ordering; its prose about result fields describes returns, not inputs.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description gives a specific verb and resource: find remote MCP servers from the official MCP Registry that Horizontal connected to and tested. Scope (tested servers only, third-party data) is precise. It does not name or differentiate from the sibling get_service, so it falls short of 5.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Usage is implied: search for servers, working ones come first, and read third-party warnings before connecting. There is no explicit statement of when to use this versus get_service, nor any when-not guidance; the level parameter's filtering intent is only hinted at in the schema.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
get_servicesearch_servicesSearch the public universe of MCP servers. This data comes from the Official MCP Registry.
Search the mcp.tc directory of MCP servers: URLs, install commands, sign-in types, tools, setup.
Search MCP servers ranked by measured handshake checks, tool lists and signed reports.
Search the official MCP registry: 17,000+ servers with trust grades, stars, tools, install config.
Add one secure layer between your agents and this server.