A side-by-side look at the Cursor Delegate and Pincushion Plugin MCP servers: transport, auth, tools, quality signals, and the config to install either one in Claude, Cursor, Windsurf or VS Code.
Verdict: choose Cursor Delegate for coding agents work as a local process; choose Pincushion Plugin for coding agents as a local process. You can also run both side by side in one mcpServers block.
delegate, cancel, doctor.get_annotations, search_annotations, get_feedback_summary.Rows where the two servers differ are marked.
| Spec | ||
|---|---|---|
| Category | Coding Agents | Coding Agents |
| Quality signal | 61/100 · Good | 64/100 · Good |
| Install command | Confirmed | Confirmed |
| Transport | Local subprocess (stdio) | Local subprocess (stdio) |
| Auth | Not declared | No auth required |
| Pricing | Not declared | Freemium |
| Required env vars | None listed | None listed |
| Verified owner | Community listing | Community listing |
| GitHub stars | 9 | 1 |
| Directory activity | 1 views 0 installs 0 upvotes | 1 views 0 installs 0 upvotes |
| Tools | 3 | 65 |
| Full listing | View Cursor Delegate | View Pincushion Plugin |
delegateDelegate a coding task to cursor-agent over ACP. Never shell out to cursor-agent — use this tool only. Auto-approves every permission the agent requests, in any mode and anywhere on disk. Clarifying questions arrive as prose in result — resume with resumeSessionId to answer. Keep model (composer-2.5), fast and effort at their defaults unless the user explicitly asks. See the delegate skill.cancelBest-effort cancel of an in-flight delegation by sessionId. session/cancel is advisory — the agent may finish the turn anyway, and the delegate result then carries cancelRequested. Status: cancelled, killed (agent process observed to exit), not-running (turn already ended, still resumable), not-found (id not in this process's recent session history).doctorSetup diagnostics: plugin version, MCP client capabilities, cursor-agent launcher resolution. Run when delegation fails or agent.found is false.get_annotationsRetrieve annotation pins from the .feedback/ directory. Filter by page URL, LWC component name, or status. Use this to understand what feedback exists before making changes.search_annotationsFull-text search across all annotation comments, selectors, component names, and tags.get_feedback_summaryGet a high-level rollup of all open feedback: counts by status, page, and component. Use this to plan what to address first.get_component_feedbackGet all feedback pins targeting a specific LWC component, with a plain-language summary ready for implementation. Returns element selectors, comments, and thread history.resolve_annotationMark an annotation as resolved after addressing the feedback. Optionally add a resolution comment explaining what was changed.add_agent_replyAdd a reply to an annotation thread (e.g. to ask a clarifying question or note a finding).get_reply_candidatesUsed by /pincushion-replies. Returns pins where Pincushion AI should respond, with each candidate tagged by trigger reason. Two triggers: (a) "mention" — the latest thread message contains @pincushion AND was authored by a human; (b) "reply-on-bot-pin" — the pin was originally authored by Pincushion AI and the latest message is from a human. Skips resolved/archived pins and any pin where the latest message is already bot-authored (idempotency). Newest-first ordering so the slash command can pace replies.add_bot_replyPost a Pincushion AI reply to a pin's thread. Hardcodes author="Pincushion AI" and authorEmail="pincushion-bot@pincushion.io" so the Chrome extension applies bot styling. ONLY call from the /pincushion-replies flow — never as a generic reply. Refuses to post if the latest thread message is already bot-authored (defense-in-depth idempotency).get_pending_critiquesUsed by /critique-latest-deploy. Lists pending critique requests queued by the deploy-hook for the current license. Newer receipts also include a deployment event, prior page outcomes, project-page inventory, and latest per-device capture receipts for truthful local reconciliation before Crit runs. Newest-first. Free on all plans (the gating happened at enqueue time on the deploy-hook side: Pro/Team — including trial-Pro — always enqueue; free licenses enqueue while their monthly AI-action pool has headroom).get_pending_deployment_frame_refreshesLists production deployment frame-refresh work independently of the Critique queue. Use this even when autoCritique is false. The result is redacted: it contains no deployment URL/hash, browser state, storage path, or capture proof. Start a selected run before capturing.start_deployment_capture_runStarts one short-lived local frame-refresh run for an exact deployment and returns its opaque proof once. Keep the proof only in process and pass it to authenticated upload_page_snapshot/record_deployment_page_result/complete_deployment_capture_run calls; never put it in capture-snapshot, shell argv, a report, pin, status message, or file.complete_deployment_capture_runRequests server-side frame-refresh completion after all exact page/device receipts have been recorded. The server rejects missing, stale, replayed, cross-deployment, or expired capture-run evidence and never treats a queue state as a current frame.Paste either block (or both) into your client config, e.g. claude_desktop_config.json or ~/.cursor/mcp.json. We only show a config when the install command is confirmed.
{
"mcpServers": {
"cursor-delegate": {
"command": "npx",
"args": [
"-y",
"cursor-delegate-mcp"
]
}
}
}
{
"mcpServers": {
"jcooley8-pincushion-plugin": {
"command": "npx",
"args": [
"-y",
"pincushion-mcp"
]
}
}
}