Head to head · Lead search · October 2026 research run

Clay vs Retriever

Clay scores 58.3 (C) on agent readiness against Retriever's 32 (F), and leads in 5 of 7 scored categories. Retriever leads on reliability and agent ergonomics. Both do lead search.

Best lead, contact and company data APIs for AI agents · All 253 leads comparisons

Which one, for what

Clay C

Best for Agents building prospect lists and enriching contacts or accounts through many providers at once.

Ahead on

  • Security & auth, 48 against 23
  • Payments & pricing, 45 against 30
  • Maintenance & community, 75 against 5
  • Transparency & trust, 61 against 49

Also in its favour

  • No incidents deducted, where Retriever loses 15 points for them

Watch for

API keys carry no scopes and act with the user's workspace access

Retriever F

Best for A sales team that wants an agent to build qualified lead tables and run outreach from Claude, Cursor or Codex.

Ahead on

  • Reliability, 45 against 40
  • Agent ergonomics, 75 against 66

Watch for

Its sequences send LinkedIn invitations, messages, visits and likes from connected accounts automatically, which LinkedIn forbids, so customers' accounts can be restricted

Why the edge

Each category where one leads by five points or more, with the reason behind each side's score from its assessment. The numbers are in the table below; this is what they stand for.

Maintenance & community Clay +70

Clay 75

clay plugin 2.35.0 on 4 November 2026, requiring CLI 1.14.0, whose tag sits on the 3 November release commit (30). 36 plugin releases from 7 August to 4 November, the last three (15 October, 3 and 4 November) changing skill content, plus weekly dated product roundups to the week of 26 October (20). Weekly dated product roundups and a support address, with the plugin's issue handling not visible (12 of 15). No official SDKs, the CLI is current (8 of 15). CLI binaries ship from GitHub releases, and the repository has no licence file (5 of 10). 30 + 20 + 12 + 8 + 5 = 75.

Retriever 5

No changelog, release notes or dated API change found, and the docs carry no dates; the blog's posts are marketing articles, not release notes (0 and 0). Support is an email address and a demo booking, with no public changelog or community channel that answers (5 of 25). Not in the official MCP registry and no official SDK (0). No package to check (0). The domain dates from 7 April 2026.

Security & auth Clay +25

Clay 48

A personal API key tied to the user's workspace access, with no scopes documented. The CLI uses OAuth per workspace (15 of 30). Functions need API or MCP access switched on one by one, workspace roles apply, and the changelog adds IP access restrictions (week of 21 September) and MCP client allowlisting (week of 26 October). CLI 1.14.0 (3 November) can hold commands that delete, send or run at scale for the user's approval, but the Claude Code plugin's hook still auto-approves most clay CLI commands, including ones that spend credits (8 of 20). Claygent web research returns untrusted text, and no injection guidance was found (3 of 15). Credit balances and a workflow-run query, with no per-call audit log found (6 of 15). security.txt valid to 31 July 2027 naming security@clay.com and trust.clay.com, re-read on 4 November, and SOC 2 Type II, ISO 27001 and ISO 42001 listed on the pricing page. No bug bounty found (16 of 20). 15 + 8 + 3 + 6 + 16 = 48.

Retriever 23

The MCP server's OAuth 2.1 issues an ordinary API key, and keys have no scopes or expiry and work as their creator in that organisation: plain revocable keys (20), less 10 because the MCP docs offer the key in the URL as ?api_key= (10 of 30). No read-only mode; destructive hints and explicit-request instructions on cancels and deletes, and the outreach toolset can be switched off (8 of 20). The tools return scraped pages, social posts and inbox replies, with no prompt-injection guidance found (0). Credit charges per run and a credits ledger by source, but no audit log documented (5 of 15). No security.txt (404), disclosure policy, bug bounty or certification found (0).

Payments & pricing Clay +15

Clay 45

No x402, MPP or L402 (0). Data credits from $0.05 and plan prices public, though on 4 November the plan cards still say $167 and $446 a month for Launch and Growth while the FAQ says $185 and $495 (15 of 20). A free plan with 100 data credits and 500 actions a month, and a page footer saying no credit card is required, while the FAQ lists Clay API access under Enterprise (15 of 20). A person signs up in a browser, and CLI sign-in opens a browser or a device code (0). 0 + 15 + 15 + 0 = 30. Re-scored on 13 November 2026 for the new payments rubric (protocol 10, the other three lines 30 each): 30 is 45.

Retriever 30

No x402, MPP or L402 (0). Credits per action and the dollar price per credit ($0.0134 on Starter) are public without a login (30). No free tier or trial without a card found; the terms say only that free tiers may be offered (0). Access needs a browser signup and a subscription, so an agent can't onboard itself (0).

Transparency & trust Clay +12

Clay 61

Closed service under the Clay Services Agreement, last updated 1 November 2026 (New York law), and a plugin repository with no licence (12 of 30). The privacy policy still dates from 11 September 2020, gives no retention periods and says nothing on AI training. The new terms incorporate a DPA with breach notice without undue delay and 30 days' notice of new sub-processors, and say customer data isn't used to train models without approval, but name no retention periods (8 of 30). Deprecated endpoints are marked in the spec without removal dates (8 of 20). The DPA links a sub-processor list through a Google Doc that points to trust.clay.com/subprocessors, a Vanta page drawn by script whose robots.txt answers with the app page, so the list wasn't read (0). 12 + 8 + 8 + 0 = 28.

Retriever 49

Closed source with clear terms, last updated 30 August 2026 (15). The privacy policy states retention (deletion within 30 days of closing an account) and names some subprocessors, but says nothing about the LinkedIn and email accounts customers connect, the people in lead tables or the data providers named on the pricing page (12 of 30). The terms say the service may be changed or removed "with reasonable notice" and material changes to the terms get 15 days' notice; no deprecation dates found (8 of 20). Subprocessors are given as examples (Vercel, Railway, Supabase, Anthropic, OpenAI) with no data locations, beyond Standard Contractual Clauses for transfers (8 of 20).

Agent ergonomics Retriever +9

Clay 66

Table queries select fields and take a limit, and searches return up to 500 per request (20 of 25). Cursor pagination and stateful iterators, structured filters and an advanced query language (20). Status codes as the error category and a message, with no stable codes in the API, though the CLI has typed exit codes (10 of 20). Items carry caller IDs for matching results and failed enrichments aren't charged, but there are no idempotency keys (8 of 20). No official SDKs, and the CLI is JSON-first (8 of 15). 20 + 20 + 10 + 8 + 8 = 66.

Retriever 75

67 tools in the catalogue (5), with 5 back because the outreach tools are a toolset shown only when the organisation has a sender and switchable in settings (10 of 25). Limit, offset and cursor pagination, up to 1,000 rows a call, and the first 20 rows returned with a finished run (18 of 20). Stable error codes with the action to take (20). Idempotency-Key on run creation, and readOnlyHint, destructiveHint and idempotentHint on the tools (20). Few required parameters (a run needs only a prompt), but no official SDK, only a Node client in a recipe (7 of 15).

Reliability Retriever +5

Clay 40

Graded with the hosted lines on the Public API. status.clay.com is linked from every page footer but answered with a Cloudflare bot check on 14 October, and on 4 November its robots.txt answered the same check, so the host is treated as blocked and its components stay unconfirmed (10 of 20, per the 14 October check, not re-read on 4 November). No readable history (5). A per-workspace request limit with no number, and published search-result caps per plan (8 of 15). 429 with Retry-After and X-RateLimit headers, backoff with jitter, and retries of 5xx only when a request is safe to retry, re-read on 4 November. No idempotency keys (12 of 15). The terms of 1 November 2026 replaced the May 2024 version and no longer set the 99.0 per cent monthly uptime target with fee credits, and no other SLA was found (0). The API is v0, and Workflows and run search are labelled alpha or beta (5 of 10). 10 + 5 + 8 + 12 + 0 + 5 = 40.

Retriever 45

No status page found, linked from neither the site nor the docs (0), so no incident history to read (5). Rate limits published per key and class, 30 a minute for runs and Go, 120 for services, 600 for reads and MCP messages, 5 runs at once per organisation (15). 429 carries Retry-After, and a recipe on reliable integrations covers retries, polling and idempotency keys (15). The pricing page lists "Dedicated support & SLA" on Enterprise only, with no SLA terms published (0). The API and MCP server aren't labelled beta (10).

Score by category

CategoryWeightClayRetrieverEdge
Reliability16%4045Retriever +5
Performance10%pendingpendingpendingnot scored in this run
Schema & documentation13%8482Clay +2
Agent ergonomics13%6675Retriever +9
Security & auth14%4823Clay +25
Payments & pricing10%4530Clay +15
Task success10%pendingpendingpendingnot scored in this run
Maintenance & community7%755Clay +70
Transparency & trust7%6149Clay +12
Negative events≤150-15
TotalC 58.3/100F 32/100

Facts side by side

FactClayRetriever
KindHTTP APIHTTP API
VendorClay Labs Inc.Retriever
Hosted endpointhttps://api.clay.com/public/v0https://api.retriever.run/mcp
TransportsHTTPStreamable HTTP
AuthOAuth or keyOAuth or key
PricingFreemiumPaid
x402nono
LicenceProprietary service under the Clay Terms of Service. The agent plugin repository carries no licence filenone
Tools exposednone67
Read-only variant documentednoyes
llms.txtyesyes
Last release2026-11-04none
Terms last updatednot read yet2026-08-30
Privacy policy last updated2020-09-112026-08-30
Customer content may train modelsnot read yetnot found in the text
Terms restrict automated accessnot read yetyes
Terms restrict benchmarkingnot read yetnot found in the text
Terms or service can change without noticenot read yetnot found in the text
Arbitration or class-action waivernot read yetnot found in the text

How the scores have moved

Verdicts

Clay

A documented Public API with an OpenAPI spec, llms.txt and plan-capped search, plus a JSON-first CLI and plugin released 36 times since 7 August, give agents Clay's people search and multi-provider enrichment. Keys have no scopes, the Claude Code plugin auto-approves most CLI calls, the new terms set no uptime target, and the privacy policy dates from 2020.

Retriever

Retriever automates outreach from customers' LinkedIn accounts, which LinkedIn's rules forbid. For agents it has an OAuth MCP server, an OpenAPI file, idempotency keys, stable error codes and published credit prices, but no status page, changelog, security contact, free tier or named company behind it.

Before you call either

Clay

  1. Fetch GET /search/query-mode/reference before writing a search query
  2. Poll routine results until a terminal 200, since 202 means still running
  3. Expect HTTP 402 when a plan's search cap is hit, and tell the user the reset date
  4. Check GET /credits/balance before large batch runs
  5. When a CLI command answers approval_required, tell the user what it would do and let them approve or deny it
  6. Ask the user before enabling the plugin's auto-approve hook in a shared workspace

Retriever

  1. Don't enrol leads in a sequence with LinkedIn steps without the user knowing LinkedIn forbids automated activity and can restrict their account
  2. Send an Idempotency-Key with each run and replay with the same key after a timeout, so a retry doesn't start a second billed run
  3. Run getEmail, getPhoneNumber and webResearch through service_batch_execute, even for one person; service_execute refuses them
  4. Check credits before a paid run, and call service_inspect for a service's thingsToKnow and credit cost before executing it
  5. On 429 wait for Retry-After; at most 5 runs execute per organisation, so a queued run starts on its own and shouldn't be created again

Questions

Which is better for AI agents, Clay or Retriever?

Clay scores 58.3 (C) on agent readiness against Retriever's 32 (F), and leads in 5 of 7 scored categories. Retriever leads on reliability and agent ergonomics.

Do Clay and Retriever need an API key?

Both take an API key or an OAuth sign-in.

Can an agent call Clay and Retriever without installing anything?

Yes. Clay has a hosted endpoint at https://api.clay.com/public/v0 and Retriever at https://api.retriever.run/mcp.

Other comparisons with Clay or Retriever

Machine-readable

For companies

Do agents find, use and choose your tools?

An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.