For AI agents · MCP + REST API
The outreach MCP server for your AI agent
Connect Claude Code, Cursor, or any MCP client. Your agent pulls warm buyers scored by buying intent, drafts in your voice, and runs your campaigns — building them, filling them, shaping what each person is sent. The one thing it can never do is approve or send. That’s yours.
One command · Claude Code
claude mcp add --transport http easeclaw https://www.easeclaw.com/api/mcp \ --header "Authorization: Bearer ec_live_YOUR_KEY"
Included with every EaseClaw plan · never posts to public feeds · cannot approve or send
Connect your agent
On claude.ai? One click below — no key to copy. For everything else, mint a key at easeclaw.com/leads/api (shown once, revoke anytime), then pick your client. Choose the outreach scope at mint time if you want your agent running campaigns as well as finding buyers.
Connect from claude.ai — one clickNo API key needed — OAuth handles it · read and draft
- In claude.ai, open Settings → Connectors → Add custom connector.
- Paste
https://www.easeclaw.com/api/mcp— leave the OAuth fields empty. - Sign in to EaseClaw and click Approve. The connection appears on your API & Agents page — revoke it there anytime.
Claude Code — one command
claude mcp add --transport http easeclaw https://www.easeclaw.com/api/mcp \ --header "Authorization: Bearer ec_live_YOUR_KEY"
Cursor — .cursor/mcp.json
{
"mcpServers": {
"easeclaw": {
"url": "https://www.easeclaw.com/api/mcp",
"headers": { "Authorization": "Bearer ec_live_YOUR_KEY" }
}
}
}CLI — npm i -g easeclaw
npm install -g easeclaw export EASECLAW_API_KEY=ec_live_YOUR_KEY easeclaw leads --min-score 70 easeclaw draft <lead-id> --tone friendly easeclaw campaigns easeclaw replies
Anything else — plain REST
curl -H "Authorization: Bearer ec_live_YOUR_KEY" \ "https://www.easeclaw.com/api/v1/leads?min_score=70&limit=10"
Eighteen tools. Find the buyer, run the outreach.
Every key gets the six find-and-draft tools. Keys minted with the outreach scope add twelve campaign tools. None of them — ever — approves or sends.
Find
list_leadsWarm buyers, best first — posts scored 40–100 by buying intent, plus LinkedIn commenters surfaced from 15 with a signal label (hand_raise, question, pain, commented).list_projectsThe businesses your agent works for: website, buyer-intent keywords, the sources it covers.Read
get_leadThe complete source post, the score and its reasoning, the current draft.get_activityRecent runs by your agent — what was scanned, what qualified.Draft
draft_replyA reply in the owner's voice — steer with tone, length, a revision note, or up to 3 variants. Uses the plan's daily AI budget.Track
update_leadWork the pipeline — new / saved / dismissed / contacted / won / lost — or write back an edited draft.Run
Twelve tools, on keys minted with the outreach scope (Pro and Business plans).
list_campaigns · get_campaignThe funnel as it stands: who is queued, invited, connected, replied — and what each person's invite and first message currently say.create_campaign · update_campaign · stop_campaignSet the play up — who it reaches, what it says, who approves it — change it later, or switch it off and recall everything still queued.get_sequence · set_sequenceRead and shape the playbook itself: the steps in order, the waits between them, the branches, and the text each one carries. Shaping it decides what the campaign will say — never that it says it.run_sourcing · add_lead_to_campaignFill it: mine a competitor's audience, resolve a pasted list, import rows, re-run a saved search, or drop one warm lead's author in. Everyone lands in the review queue.update_prospectShape one person: edit exactly what will go out for them, record a booked meeting, or take them out before anything is committed.list_replies · get_sender_statusWho wrote back, newest first — and whether a LinkedIn sender is connected to carry out what you approve.Send
No tool exists for this step. Approving what goes out, and sending it, stay with you and the limits you set — from your dashboard, or inside your campaign’s own autopilot limits. Public replies your agent drafts and you post yourself. That’s the product rule, not a setting.
Prefer plain HTTP?
The same key works on the REST API — for scripts, n8n, Zapier webhooks, anything. Full schema in the OpenAPI spec.
| GET /api/v1/leads | List warm leads — project_id, status, min_score, since, limit |
| GET /api/v1/leads/:id | One lead in full |
| PATCH /api/v1/leads/:id | Update status and/or write back an edited draft_reply |
| POST /api/v1/leads/:id/draft | Generate a reply draft — tone, length, note, n |
| GET /api/v1/projects | The businesses your agent works for |
| GET /api/v1/activity | Recent runs by your agent — what was scanned, what qualified |
| GET · POST /api/v1/campaigns | Your campaigns with funnel counts; create one |
| GET · PATCH /api/v1/campaigns/:id | One campaign in full; change its settings |
| GET · PUT /api/v1/campaigns/:id/sequence | The playbook it runs — read it, or replace it |
| POST /api/v1/campaigns/:id/stop | The off switch — recalls everything still queued |
| POST /api/v1/campaigns/:id/prospects | Put one lead's author into the review queue |
| PATCH …/prospects/:pid | Edit one person's message, mark a meeting, or take them out |
| POST /api/v1/campaigns/:id/source | Fill it — competitor audience, pasted list, CSV, saved search |
| GET /api/v1/inbox/replies | Everyone who wrote back, newest first |
| GET /api/v1/senders | Whether a LinkedIn sender is connected |
Bearer auth · 300 requests/hour per account · drafting capped by your plan’s daily AI budget · campaign endpoints need a key minted with the outreach scope · no endpoint approves or sends.
Questions
What is an MCP server?
MCP (Model Context Protocol) is the open standard AI agents use to talk to tools. Adding EaseClaw's MCP server gives your agent — Claude Code, Claude Desktop, Cursor, or any MCP client — direct access to your pipeline: it can list warm buyers, read the source posts, draft in your voice, update pipeline status, and (with an outreach-scope key) build, fill and shape the campaigns that reach them.
Can my agent post or send replies for me?
Never. There is no tool on this server that posts, sends, or approves — on either roster. Your agent prepares the work; you post every public reply yourself. An outreach-scope key adds the campaign tools, and even those cannot approve a single message: invites and first messages leave only after your approval, or inside the autopilot limits you set. That's a hard product rule, not a setting.
What does it cost?
The MCP server and REST API are included with every EaseClaw plan ($99 or $199/mo, and every account starts with a free 7-day trial that takes no card). Reply drafting uses your plan's daily AI budget, which is hard-capped, so an agent loop can never run up a surprise bill.
Which tools does my agent get?
Every key gets six: list_projects, list_leads, get_lead, draft_reply, update_lead, get_activity — find the buyer, read the post, draft in your voice, move it through your pipeline. A key you mint with the outreach scope adds twelve more: list_campaigns, get_campaign, create_campaign, update_campaign, get_sequence, set_sequence, stop_campaign, add_lead_to_campaign, run_sourcing, update_prospect, list_replies, get_sender_status. Eighteen in all, and not one of them approves or sends — get_sequence and set_sequence shape the playbook a campaign runs, which is what it will say once you approve it or your limits allow it.
Does the claude.ai connector get the campaign tools?
No, and it never will. The one-click connector's consent screen promises it can never post or send anything for you, so a connector key stays read-and-draft for its whole life — that is enforced in the database, not by convention, and a connector session is never even shown a campaign tool. The outreach scope exists only on keys you mint yourself in the dashboard, where you choose it deliberately.
How is it secured?
Two ways in. Mint an API key in your dashboard (only a hash is stored — the key is shown once), or connect claude.ai in one click via OAuth — the authorization appears as a connected app on your API & Agents page. Requests are rate-limited to 300/hour per account, drafting is budget-capped, and you can revoke any key or connected app at any time — anything using it stops working immediately.
How is this different from other lead tools' MCP servers?
Most agent integrations in this space stop at reading mentions or scoring posts. EaseClaw's MCP closes the loop: the same connection that finds and scores the buyer also drafts in your voice, builds and fills the campaign that reaches them, shapes what each person is sent, and reports who wrote back. Everything except approving and sending, which stay with you and the limits you set.
Give your agent a pipeline.
Start the free trial, mint a key, and your agent is pulling scored leads two minutes from now.