# plyce MCP

Endpoint: https://mcp.plyce.app/mcp
Docs: https://mcp.plyce.app/docs/capabilities.md
Scopes (read): mcp candidates:read projects:read companies:read contacts:read activities:read reporting:read mcp:manage

## Can
- get_current_user_context: Return the logged-in plyce user, tenant, scopes, and which tools they may call.
- get_allowed_values: Enums the agent must use (candidate status, pipeline status, activity types).
- get_mcp_connection_status: Show this MCP connection (client, scopes, last seen).
- get_mcp_usage: Show remaining quota for this user in the current window.
- revoke_mcp_connection: Revoke this MCP connection immediately.
- search: Database search over candidates, projects, companies, contacts the user can see.
- fetch: Load one allowlisted record by composite id from search.
- get_candidate: Allowlisted candidate fields plus a plyce deep link.
- get_project: Allowlisted project fields plus a plyce deep link.
- get_company: Allowlisted company fields plus a plyce deep link.
- get_contact: Allowlisted contact fields plus a plyce deep link.
- get_activity_report: Deterministic KPI counts (including contact.sales_call) for a Europe/Berlin period. Same definitions as plyce reporting.
- list_report_activities: List matching activity rows for a report period. Drill-down still requires object rights.
- get_pipeline_report: Counts of Sent out / Interview / Offer on active projects.
- get_time_report: KPI time buckets in Europe/Berlin (not plyce time timesheet approval).
- list_open_tasks: List open manual tasks (mail_automation_scheduled_emails variant task).
- get_daily_briefing: Structured list of open tasks due today/overdue. No plyce-generated prose.
- get_entity_360: Allowlisted record plus recent activity titles (no note bodies from other users beyond title).

## Cannot
- No plyce AI: matching, CV analysis, job-ad analysis, generated email/dossier/briefing prose.
- No SQL, no service-role data path for CRM reads.
- No salary, gender, birth date, CV, or dossier fields.
- No mail send from MCP. prepare_email stores a draft; send is confirmed in plyce.
- No pipeline status or timesheet decision commit from MCP.
- No token passthrough of the plyce browser session.
- get_current_user_context: Not a way to impersonate another user or list other tenants.
- get_allowed_values: Not a search and not a schema dump of the database.
- get_mcp_connection_status: Not an admin dump of every connection in the tenant.
- get_mcp_usage: Not billing and not other users’ usage.
- revoke_mcp_connection: Does not delete CRM data.
- search: Does not rank against a job ad, does not call plyce matching, does not read CVs. The user model ranks locally.
- fetch: Not a SQL proxy and not a document/CV download.
- get_candidate: No salary, gender, birth date, CV text, or dossier.
- get_project: No salary band or full job-ad HTML (user model must not be fed raw ads as a matching engine).
- get_company: Not news scraping or plyce-AI company research.
- get_contact: Not an email send.
- get_activity_report: Does not write prose. Does not call kpi-calculate LLM paths.
- list_report_activities: Not a full activity dump across tenants.
- get_pipeline_report: Does not change pipeline status.
- get_time_report: Does not approve timesheets or create invoices.
- list_open_tasks: Does not complete or send mail.
- get_daily_briefing: Not an LLM summary and not matching.
- get_entity_360: Not a CV or document viewer.
- add_note: Does not generate the note text. plyce only stores it. One canonical activity row.
- create_task: Does not email the assignee and does not log an activity until the task is completed in plyce.
- set_candidate_tags: Does not change pipeline auto-tags.
- update_candidate_availability: Does not run status automation beyond normal DB triggers.
- prepare_email: Does not call plyce-AI to write the mail. Does not send. Does not invoke email-send.

You are connected to plyce, a recruiting CRM. Data is limited to the signed-in user and their tenant (RLS). plyce does not run an LLM on this MCP. Ranking, summaries, and email wording are your job from structured fields. Read tools: get_current_user_context, get_allowed_values, get_mcp_connection_status, get_mcp_usage, revoke_mcp_connection, search, fetch, get_candidate, get_project, get_company, get_contact, get_activity_report, list_report_activities, get_pipeline_report, get_time_report, list_open_tasks, get_daily_briefing, get_entity_360. Write/prepare tools exist only when enabled for the tenant: add_note, create_task, set_candidate_tags, update_candidate_availability, prepare_email. If a tool is missing from tools/list, do not invent it. Never match candidates to a job ad via plyce. Use search + your own ranking. No CV text, salary, gender, or dossier. Do not send email. prepare_email stores a draft; a human confirms send in plyce. Deep-link users back to plyce URLs returned by get_* tools. Timezone for "yesterday" / "today" is Europe/Berlin. Sales calls are event_key contact.sales_call. Forbidden: analyze_job_ad, search_candidates_for_job_ad, match_candidates_to_project, explain_candidate_match, find_similar_candidates, compare_candidates, get_daily_briefing_prose, generate_email, generate_outreach.