MCP Server Search
Searches public Model Context Protocol servers by capability or name and returns each with its description, repository, hosted endpoint or package, and version.
- Price per call
- 1 credit
- per successful call
- Price ceiling
- 1 credit
- every call, same price
- Failed calls
- Free
- retried automatically
- Billed
- Per call
- only on success
Endpoint
api.primaryapi.com/ v1/ run/ mcp. searchAuth: API key with run or full access, or a dashboard session
Send the input below as a JSON body. The call is routed to a healthy upstream; if an attempt fails, it is retried automatically inside the same request and you pay only for the answer you get. To check an input and see the exact price without running anything, call POST /v1/quote/mcp.search with the same body (free).
Example request
curl -sS -X POST "https://api.primaryapi.com/v1/run/mcp.search" \
-H "Authorization: Bearer $PRIMARYAPI_API_KEY" \
-H "Content-Type: application/json" \
-d '{"query":"postgres"}'const res = await fetch('https://api.primaryapi.com/v1/run/mcp.search', {
method: 'POST',
headers: {
Authorization: `Bearer ${process.env.PRIMARYAPI_API_KEY}`,
'Content-Type': 'application/json',
},
body: JSON.stringify({
query: 'postgres',
}),
});
const body = await res.json();
if (!res.ok) throw new Error(`${body.code}: ${body.error}`);
// found:false is a successful answer ("does not exist"), not an error
console.log(body.output.found ? body.output.data : body.output.reason, body.credits);import os
import requests
res = requests.post(
"https://api.primaryapi.com/v1/run/mcp.search",
headers={"Authorization": f"Bearer {os.environ['PRIMARYAPI_API_KEY']}"},
json={
"query": "postgres",
},
timeout=60,
)
body = res.json()
if not res.ok:
raise RuntimeError(f"{body['code']}: {body['error']}")
# found=False is a successful answer ("does not exist"), not an error
print(body["output"]["data"] if body["output"]["found"] else body["output"]["reason"], body["credits"]){
"name": "run_api",
"arguments": {
"sku": "mcp.search",
"input": {
"query": "postgres"
}
}
}Price this exact input first (free)
curl -sS -X POST "https://api.primaryapi.com/v1/quote/mcp.search" \
-H "Authorization: Bearer $PRIMARYAPI_API_KEY" \
-H "Content-Type: application/json" \
-d '{"query":"postgres"}'Returns credits (what the call is expected to cost) and maxCredits (the most it can cost, which is set aside while it runs). Invalid input fails here the same way it would on run.
Input
JSON body fields. Unknown fields are rejected with 400 invalid_input (never silently ignored). You may add the routing controls preferLatencyUnderMs and maxCredits to the same object.
querystringrequired- What the server should do or be called, e.g. "postgres" or "github issues" (3+ characters).max 100 chars
remoteOnlybooleandefaultfalse- Only servers with a hosted (remote) endpoint.
includeDeprecatedbooleandefaultfalse- Also return servers their publisher marked deprecated.
limitintegerdefault20- Maximum number of results to return (1-50, default 20).1–50
Output
When output.found is true, output.data has these fields (a list under servers, billed per item returned).
serversobject[]- Best matches first, then by name.
servers[].namestring- Reverse-DNS server name, e.g. "io.github.owner/server".
servers[].titlestring | nullservers[].descriptionstring | nullservers[].versionstring | nullservers[].repositorystring | null- Source repository URL.
servers[].websiteUrlstring | nullservers[].remotesobject[]- Hosted endpoints you can connect to.
servers[].remotes[].typestring- streamable-http or sse.
servers[].remotes[].urlstringservers[].packagesobject[]- Installable packages (run locally).
servers[].packages[].registryTypestring- npm, pypi, oci, nuget, mcpb, …
servers[].packages[].identifierstringservers[].packages[].versionstring | nullservers[].packages[].transportstring | nullservers[].statusstring- active or deprecated.
servers[].publishedAtstring | nullservers[].updatedAtstring | nullservers[].scoreinteger- Query words found in name, title or description (higher ranks first).
totalinteger- Matching servers before `limit`.
nextCursorstring | null
TypeScript type
/** output.data when output.found is true */
type McpSearch = {
/** Best matches first, then by name. */
servers: {
/** Reverse-DNS server name, e.g. "io.github.owner/server". */
name: string;
title: string | null;
description: string | null;
version: string | null;
/** Source repository URL. */
repository: string | null;
websiteUrl: string | null;
/** Hosted endpoints you can connect to. */
remotes: {
/** streamable-http or sse. */
type: string;
url: string;
}[];
/** Installable packages (run locally). */
packages: {
/** npm, pypi, oci, nuget, mcpb, … */
registryType: string;
identifier: string;
version: string | null;
transport: string | null;
}[];
/** active or deprecated. */
status: string;
publishedAt: string | null;
updatedAt: string | null;
/** Query words found in name, title or description (higher ranks first). */
score: number;
}[];
/** Matching servers before `limit`. */
total: number;
nextCursor: string | null;
};If what you asked about does not exist, is private, or a search matches nothing, the call still succeeds: output.found is false, output.data is null and output.reason says why. Such misses are free unless the notes below say misses are billed, in which case a miss costs the price of one call.
Example response
Values are placeholders that follow the output schema (strings as "string", numbers as 0); credits shows what this call would cost at the usual price.
{
"output": {
"found": true,
"data": {
"servers": [
{
"name": "string",
"title": "string",
"description": "string",
"version": "string",
"repository": "string",
"websiteUrl": "https://example.com",
"remotes": [
{
"type": "string",
"url": "https://example.com"
}
],
"packages": [
{
"registryType": "string",
"identifier": "string",
"version": "string",
"transport": "string"
}
],
"status": "string",
"publishedAt": "string",
"updatedAt": "string",
"score": 0
}
],
"total": 0,
"nextCursor": null
}
},
"provider": "PrimaryAPI",
"credits": 1,
"items": 1,
"requestId": "req_01JBXK6Y2N9V8T4R3Q2P1M0A7Z",
"nextCursor": null
}Pricing and reliability
Calls go only to upstreams that are healthy right now. If an attempt fails, the call is retried automatically inside the same request; a failed attempt costs nothing, and a call never costs more than its quote’s maxCredits. Live figures below are measured on real traffic.
Success rate, uptime, latency and call counts are measured live from real traffic over 30 days and show “—” until there is data.
Errors
Errors are JSON { code, error, requestId } and are never charged. The ones to handle for this endpoint:
- invalid_inputThe request failed validation before anything ran.
- unauthorizedNo valid API key or session was presented.
- insufficient_balanceThe balance cannot cover the most this call could cost.
- key_limit_reachedThe key’s own spend limit is used up for the current window.
- trial_cap_reachedAn agent trial key has used its whole trial budget.
- forbiddenThe credential is valid but may not do this.
- not_foundThe thing addressed by the URL does not exist for you.
- idempotency_conflictThe
Idempotency-Keyis already bound to a different request. - rate_limitedToo many requests in the current window.
- all_sources_failedThe call could not be answered, even after automatic retries.
- service_unavailableThis API cannot run in this deployment right now.
- timeoutThe request passed its deadline before it finished.
Notes
Candidates are servers whose name contains your query or one of its words ("postgres" finds "io.github.x/postgres-mcp"); they are ranked by how many of your words appear in the name, title and description. Only the latest version of each server is returned. No match is a free miss (reason no_results). Listings are metadata published by server authors: evaluate a server before you connect it.
Generated from the PrimaryAPI catalog. Machine-readable contract: GET https://api.primaryapi.com/v1/apis/mcp.search.