list_projects
scope: readfreeList all projects in the authenticated workspace, including competitors and target ICP.
No parameters. Send an empty JSON object.
Agent reference
Every tool is defined once in @launchengine-dev/contracts and served two ways: as an MCP tool, and over HTTP. The schemas below are read from that registry, so they match what the server accepts.
Authentication
Create a key on the agents page, then send it as a bearer token:
http
Authorization: Bearer le_live_...
A key is scoped to one workspace. Requests are rate limited per key (default 60/minute); exceeding it returns a retryable error.
MCP server
The MCP server is a thin proxy to the same endpoints, published as @launchengine-dev/mcp. It needs no clone and no local API — the default points at production.
claude_desktop_config.json
{
"mcpServers": {
"launchengine": {
"command": "npx",
"args": ["-y", "@launchengine-dev/mcp"],
"env": { "LAUNCHENGINE_DEV_API_KEY": "le_live_..." }
}
}
}
Claude Code needs one command instead:
bash
claude mcp add launchengine \
--env LAUNCHENGINE_DEV_API_KEY=le_live_... \
-- npx -y @launchengine-dev/mcp
Running npx -y @launchengine-dev/mcp by hand prints one line and then waits — that is a working stdio server, not a hang. It answers --help and --version if you want to check the install.
Endpoint
http
POST /api/tools/<tool_name>
Content-Type: application/json
{ ...tool input... }
The response is always { ok: true, data } on success or { ok: false, error: { code, message, retryable } } on failure, using the codes in the shared contract.
Tools
get_credit_balance
scope: readfreeReturn the workspace credit balance and current plan. Free; use before a metered call.
No parameters. Send an empty JSON object.
find_intent_signals
scope: read1 creditScan Reddit/X/HN (via SERP indexes) for high-intent posts matching a project's keywords and competitors. Returns scored signals. Metered.
| Field | Type | Required | Default |
|---|---|---|---|
| project_id | string | yes | — |
| min_intent_score | number | no | 60 |
| platforms | array | no | ["hackernews","reddit","web"] |
| since_days | number | no | 30 |
| status | enum | no | — |
| limit | number | no | 20 |
| scan | boolean | no | false |
| web_mode | enum | no | "single" |
| include_muted | boolean | no | false |
get_intent_details
scope: readfreeFetch full details for a single intent signal by id, including the post snippet.
| Field | Type | Required | Default |
|---|---|---|---|
| intent_id | string | yes | — |
draft_reply
scope: write1 creditDraft a consultative reply to an intent signal. Assistive only: a human must review and post it. Metered.
| Field | Type | Required | Default |
|---|---|---|---|
| intent_id | string | yes | — |
| tone | enum | no | "consultative" |
subscribe_to_keyword
scope: writefreeCreate or update a monitoring subscription for a keyword and competitor set within a project.
| Field | Type | Required | Default |
|---|---|---|---|
| project_id | string | yes | — |
| keyword | string | yes | — |
| competitors | array | no | [] |
Credits
Metered tools deduct credits before the work runs and refund on failure. Call get_credit_balance before a metered call if you need to check first — it is free.
Assistive drafting
draft_reply always returns requires_human_post: true. Nothing is posted on your behalf, by design — Reddit and Hacker News treat automated self-promotion as spam. generated_by reports whether a model or the deterministic fallback produced the text, so a caller can tell a real generation from a generic one.