Quickstart: your first PumpGTM API and MCP calls

Everything PumpGTM does is available to code and to AI agents. This page gets you from zero to a working call. You need a PumpGTM workspace (7 day trial at app.pumpgtm.com/onboarding) and one connected LinkedIn account.

1. Get your workspace key

Sign in at app.pumpgtm.com, open MCP in the left nav, and copy the workspace key. It starts with eve_mcp_. One key is one workspace: it can read and act on that workspace's leads, sequences and replies and nothing else. Keep it server side.

export PUMPGTM_KEY="eve_mcp_..."

2. First REST call: what happened this week

curl -s "https://app.pumpgtm.com/api/v1/progress?sinceDays=7" \
  -H "Authorization: Bearer $PUMPGTM_KEY"

You get the funnel (invited, accepted, messaged, replied, meetings booked), the queue depth, and every connected LinkedIn account with today's connection-request budget. Full shape in the REST API reference.

Three more calls worth trying:

# Every sequence with its steps, status and lead count
curl -s https://app.pumpgtm.com/api/v1/sequences -H "Authorization: Bearer $PUMPGTM_KEY"
 
# Connection requests sent in the last 7 days (read `count`)
curl -s "https://app.pumpgtm.com/api/v1/activity?action=invite&outcome=ok&sinceDays=7&limit=1" -H "Authorization: Bearer $PUMPGTM_KEY"
 
# Replies waiting for a human, each with a drafted answer
curl -s "https://app.pumpgtm.com/api/v1/replies?limit=5" -H "Authorization: Bearer $PUMPGTM_KEY"

Generating a typed client: the spec is at pumpgtm.com/openapi.json (OpenAPI 3.1).

3. First MCP call: connect an agent

The MCP server is hosted at https://mcp.pumpgtm.com/mcp (Streamable HTTP). The same key authenticates it.

Claude Code:

claude mcp add --transport http pumpgtm https://mcp.pumpgtm.com/mcp \
  --header "Authorization: Bearer $PUMPGTM_KEY"

Codex:

codex mcp add pumpgtm --url https://mcp.pumpgtm.com/mcp && codex mcp login pumpgtm

Cursor and any other client: add a server with that URL and the Authorization: Bearer header. Claude (web, desktop) and ChatGPT: add the URL as a connector and sign in; they use OAuth instead of the key.

Then, in the agent:

Call get_workspace and tell me what is waiting on me.

get_workspace returns setup state, sequences, unfinished reviews and pending replies, plus a next hint for the following call. The 24 tools are listed in the MCP reference, with the live schema at pumpgtm.com/mcp/tools.json.

4. A first real workflow, safely

Nothing reaches LinkedIn until a person approves it, so you can run this end to end without contacting anyone:

1. find_people with targeting and confirmed: false. It saves a Play and returns the targeting summary.
2. find_people with the returned playId and confirmed: true. It runs discovery and returns candidates.
3. review_people with decisions and finish: true. Good fits are queued into a sequence. Still nothing sent.
4. save_sequence_draft to write the steps, set_sequence_status approved with the exact revision, then active. Now outreach starts, paced inside LinkedIn's limits.
5. Replies freeze the lead. Read them with list_pending_replies or GET /api/v1/replies, decide with decide_reply or POST /api/v1/replies/{id}/decide.

5. Building on top of PumpGTM

• Your own dashboard: poll /api/v1/progress, /api/v1/reports/funnel, /api/v1/leads and /api/v1/activity. Counts are exact and every timestamp is UTC.
• Your own reply handling: subscribe a webhook to reply.received (POST /api/v1/webhooks), or poll GET /api/v1/replies; decide is the action. The same code runs behind the in-app buttons, so a human and your integration can pick up where the other left off.
• Agencies and platforms with many client workspaces: one partner key, one MCP URL per end user, usage and caps by API. See pumpgtm.com/mcp/platforms.
• Anything an agent reads here is also plain Markdown: add .md under /markdown/pages/docs/ or request any page with Accept: text/markdown. The site index for agents is llms.txt.

Questions or a missing endpoint: hello@pumpgtm.com.