---
name: agentsearch-find-agent
description: Find, vet and browse AI agents and MCP servers through Luther Systems' AgentSearch — public, free, no auth — over HTTP, MCP or A2A.
api: openapi/luthersystems-com-agentsearch-openapi.yml
operations: [searchAgents, getAgent, scoreFoundAgent, browseAgents, getStats]
mcp: https://agentsearch.luthersystems.com/api/mcp
a2a: https://agentsearch.luthersystems.com/api/a2a
generated: '2026-09-19'
method: generated
source: Generated by API Evangelist from openapi/luthersystems-com-agentsearch-openapi.yml, the live MCP tools/list, https://agentsearch.luthersystems.com/api-docs and /llms.txt. Not provider-published.
---

# Find an agent with AgentSearch

AgentSearch (agentsearch.luthersystems.com) is a daily-scored index of ~4,300 AI agents and MCP
servers. Every surface is public, needs no key, and is rate-limited only on a best-effort basis
(~60 requests per minute per IP). Pick the surface you already speak:

- **MCP** — `POST https://agentsearch.luthersystems.com/api/mcp` (Streamable HTTP), or
  `npx -y @luthersystems/agentsearch` for stdio. Tools: `search`, `agent_details`, `found_agent`,
  `browse`, `stats`.
- **HTTP** — the operations below, base `https://agentsearch.luthersystems.com`.
- **A2A** — JSON-RPC `message/send` at `/api/a2a`; skills carry the same five ids.

## 1. Search by what you need done — `searchAgents`

`POST /api/search` with `{"query": "read invoice PDFs", "limit": 5}`. Results are `SearchResult`
objects ranked by `final_score = query_fit × overall_score`, each with a one-line `reasoning`.
Treat `query_fit` (does it match the task) and `overall_score` (is it any good) separately.

## 2. Vet one candidate — `getAgent`

`GET /api/agent/{name}` with the exact indexed name (URL-encode slashes, e.g.
`upstash%2Fcontext7`). Returns `AgentDetails`: `blocker.status` (alive/dead/timeout/error),
layer scores (fame, usability, functionality, call_graph, overall) and the raw source payload.
A **404** means the agent is not in the index — go to step 3, do not retry.

## 3. Score something not yet indexed — `scoreFoundAgent`

`POST /api/found-agent` with `{"url": "...", "query": "optional task"}` runs the same crawl,
probe and rubric live. This does **not** add the agent to the index; the provider documents
`POST /api/submit-agent` for that (it is not in the OpenAPI or the tool list).

## 4. Walk or summarise the whole index — `browseAgents`, `getStats`

`GET /api/browse?page=1&page_size=100` pages every agent by overall score;
`GET /api/stats` returns totals, source and license breakdowns and the call graph.

## Rules

- No authentication anywhere; do not send credentials.
- All operations are reads; nothing here is idempotency-sensitive or reversible-sensitive.
- Only `getAgent` declares a non-2xx (404). Other failures are undocumented — back off on 429/5xx.
- Scores come from an LLM-judged rubric (methodology: github.com/luthersystems/agentsearch/blob/main/docs/methodology.md); surface the `reasoning` to the user rather than the bare number.
