[ 01 ]
Context-switching breaks the loop
Drafting a pitch, checking a domain, or looking up a lead means leaving the editor for a dashboard — every hop costs the agent its train of thought.
CogniLead
ProductWarmed sender poolReputation circuit-breakerJurisdiction routingREST APIMCP serverDNS automationCompliance & suppression
SolutionsDevelopersAgenciesB2B SaaSRecruitersSales teamsSolo foundersSaaSRecruiting & staffingReal estateAgencies & consultingEcommerce & DTCB2B services
CompareAll 12 alternatives →vs. Instantly.aivs. Smartleadvs. Lemlistvs. Apollo.iovs. Woodpecker.coInstantly vs. SmartleadSmartlead vs. LemlistApollo vs. LemlistSee all 8 →
ResourcesBlogDocsMCP serverAPI referenceSPF checkerDMARC checkerBlacklist checkerSpam-word checkerStatus
PricingLog inMCP server
CogniLead, callable from inside your AI IDE.
@cognilead/mcp-server publishes the deliverability engine as sixteen Model Context Protocol tools — draft a pitch, score a signal, and check a domain's deliverability with no key at all. Hand it a key and the real outbound pipeline opens up, right where your agent already works.
The problem
An agent that can only chat is not an agent that can ship outbound.
[ 02 ]
Agents that can only talk, not act
A chat window that drafts good copy but cannot ingest a lead, check deliverability, or suppress an address is still just a chat window.
[ 03 ]
Brittle, hand-wired tool-calling glue
Wiring an LLM to your own API by hand means maintaining fragile request/response plumbing for every action, rewritten per project.
[ 04 ]
No way to try it before committing to a key
Most integrations demand a credential up front, before you can even tell whether the tool surface is worth wiring in.
How it works
Keyless to start. Keyed for the pipeline.
One running process, one config block — what it exposes just grows as you hand it credentials. Nothing about the tool surface changes shape; more of it just answers.
[ 01 ]
Four tools work immediately
draft_pitch, score_signal, check_deliverability, and health all answer with no signup, no key, and no expiry — an agent can learn the surface before you commit to anything.
[ 02 ]
Add a key, twelve more unlock
ingest_lead and suppress touch the real pipeline; create_topup_link and read-only tools like list_leads and get_billing_usage read your account state back.
[ 03 ]
Resources and Prompts add context
Read-only Resources (cognilead://sends/recent, cognilead://suppressions) feed live state into the model; Prompts like cold-outbound-sequence and deliverability-triage encode the playbooks.
[ live ]
Same process, more surface
A generic MCP-compatible client calling a keyless tool and a gated one, side by side.
See it called
Every call still runs through the same guardrails.
An agent gets the same tool surface a human gets through the dashboard or REST API — nothing about calling it from an IDE skips suppression or reputation checks.
> cognilead.health()free
✓{ ok: true }4ms
> cognilead.check_deliverability({ domain: 'acme-outbound.example' })free
✓SPF pass · DMARC pass · grade A61ms
> cognilead.score_signal({ signal: {...} })free
✓fit_score: 82 · recommendation: "pitch"9ms
> cognilead.draft_pitch({ signal: {...} })free
✓subject + body drafted · technical_hook_verified: true340ms
> cognilead.ingest_lead({ email: 'lead@acme-outbound.example' })gated
✓persisted · suppression: clear118ms
> cognilead.suppress({ email: 'opt-out@example.com', reason: 'unsubscribed' })gated
✓added to suppression list47ms
[ 01 ]
16
MCP tools
[ 02 ]
4
Keyless, forever
[ 03 ]
3
IDEs supported
Sixteen tools, one process
Your agent's next call is one config block away.
Four tools answer with no key. Twelve more, plus live Resources and reusable Prompts, unlock the moment you hand it one — the same guardrails either way.
16 tools total4 keyless, 12 gated4 Resources · 3 Prompts
FAQ
Questions about the MCP server.
16 total — 4 keyless, 12 gated — plus 4 Resources and 3 Prompts. The full table, with every tool name, cap, and description, lives at /mcp.
Yes, once given an API key — ingest_lead and the suppression tools run through the same intersect stage and reputation checks as the REST API. There is no separate, laxer path for an agent.
npx -y @cognilead/mcp-server, configured in your IDE’s MCP config file. Cursor, Claude Desktop, and Windsurf are all covered with copy-paste config blocks at /mcp.
draft_pitch, score_signal, check_deliverability, and health work immediately with zero configuration — enough to see the surface work before you decide whether to wire in a key.
This page covers why the MCP surface matters — for the complete tool/Resource/Prompt reference and IDE install instructions, see the full MCP tool reference and install guide.
Your leads deserve to reach an inbox.
Every email you send without protection is a gamble on whether it reaches the inbox at all. Bring your leads — CogniLead handles the warmup, the safety checks, and the cleanup, so you don't have to think about deliverability again.
100 sends a month, free forever · no credit card · cancel any time