Build on Prscnt

Already run your own agency tools — a custom manager portal, a no-code app, an internal agent? Use Prscnt as the orchestration and intelligence layer on top of the stack you already have. Keep your portal, your database, and your connectors. Plug our MCP in next to them.

This page is the quick tour. For the complete, product-agnostic developer documentation — quickstart, auth, composing any stack, the runner, the plugin, a full tool reference, and recipes — see the developer docs.

The model

Prscnt builds zero third-party API clients and stores zero third-party credentials. When work needs to touch your tools, Prscnt returns a proposed structure (a Drive folder tree, a calendar event, an invoice payload) and your agent drives your connector to execute it. Prscnt holds only thin pointers back to your records (a Prscnt deal linked to a record id + URL in your system) plus its own proprietary intelligence: rate corpus, brand reliability, partnership graph, creator-brand fit, voice profiles, deal pipeline, contract playbooks, invoicing/AR, and content analysis. Your customer data and credentials never leave your stack.

One thing to know up front. There is no plain-REST endpoint. Every Prscnt tool is reached only through an MCP client with an LLM in the loop (JSON-RPC at POST https://mcp.prscnt.com/v1/mcp). A raw Softr button or Airtable automation can't call Prscnt directly — the pattern is always your portal → a small LLM agent runner (holding the Prscnt MCP + your own connectors) → writes results back into your data.

Integration postures

a. In-Claude composition — fastest, zero code

When: your managers already work in Claude day to day. How: each manager connects their own stack MCPs (Gmail, Drive, QuickBooks, Apollo, Airtable, Monday, HubSpot) plus the Prscnt MCP in the same session. Claude selects tools across all of them in one conversation — pull a rate with rate_benchmark, draft with voice_draft, create the Gmail draft via the manager's own Gmail connector. Zero code; the human approves every outbound action.

b. Custom portal + agent backend — the headline case

When: you run a custom manager portal (e.g. Softr on Airtable, a no-code app, or an internal agent) and want Prscnt as the brain behind it.

Softr UI    managers click "Quote this deal", "Vet brand", "Draft outreach"
   |  (row state, button triggers)
Airtable    Deals, Contacts, Brands, Partnerships, Payments tables
   |  read row   ^ write results back
Orchestration agent  =  LLM runner + Prscnt MCP + Airtable MCP
                        (+ Gmail / QuickBooks / Apollo MCPs)

The agent reads an Airtable row, calls Prscnt tools, and writes results back into Airtable so they surface in Softr. Per stage:

Price a deal — brand_resolve → rate_benchmark, optionally deals_benchmark + comparator_pay_history → write recommended/floor/ceiling to the Deal row.
Vet the brand — brand_reliability, brand_fingerprint, negotiation_intel → write a reliability summary field.
Pick the creator — roster_match_brand or creator_brand_fit; check_exclusivity_conflict before committing.
Draft in-voice — voice_draft returns profile + exemplars + phrases; compose, then create the draft via the manager's own Gmail connector. Human approves before send.
Advance the deal — deals_update_stage (records history); deals_bulk_create for idempotent upserts.
Invoice + chase — invoice_prepare builds line items + commission split + a QuickBooks payload → run the user's QBO MCP to create it → invoice_record_qbo → invoice_payment_reconcile → ar_status for aging.
Redline a contract — contracts_run_playbook; contracts_get_playbook / contracts_set_playbook to view or tune.
Keep Airtable in sync — crm_import pulls your brand-contact graph in read-only; stack_link_deal points a Prscnt deal at the Airtable record; stack_resolve_deal reverse-looks-up; stack_deal_refs lists all links.

You build the Softr UI, the Airtable schema, and the agent runner glue. You get all intel, pipeline, voice, invoicing, and contract logic for free.

Start from the template. You don't have to write the runner from scratch. The complete orchestration-runner template is in the docs — five short files you copy into your own project: a webhook server (POST /run) wired to the Prscnt MCP, with slots for your own connectors and a built-in human-approval gate. Point your Softr button or Airtable automation at it and you're orchestrating. For the full skill and subagent set, install the prscnt plugin (claude plugin add) and load it via the Agent SDK plugins option — same brain, maintained by us, no rebuild on your side.

c. Programmatic / headless

No REST API or webhook returns intel JSON. To trigger Prscnt from your portal without a human in chat, stand up a small LLM agent runner (e.g. an Agent SDK process) that your Softr button or Airtable automation pings via your own webhook. That runner holds the Prscnt MCP connection + a bearer token, takes a row id, runs the sequence above, and writes back to Airtable. (You can hand-roll a raw JSON-RPC client — initialize → read Mcp-Session-Id → tools/call — but then you own all tool selection; treat as advanced.) Either way it consumes your workspace's tier rate limit and monthly allowances.

d. Event recipes

When the portal knows an event happened ("deal signed", "invoice overdue") and wants the next-best action sequence: workflow_describe lists definitions; workflow_propose returns proposed actions for user approval (never auto-fired); orchestration_recipes returns cross-tool recipe cards. Your agent surfaces the proposed steps in your portal and executes only what a manager approves.

Which Prscnt tool for which job

Job	Tool(s)	Type
Price a deal	`rate_benchmark`, `deals_benchmark`, `comparator_pay_history`	read
Vet a brand's reliability	`brand_reliability`, `brand_fingerprint`, `negotiation_intel`	read
Pick the right creator	`roster_match_brand`, `creator_brand_fit`, `check_exclusivity_conflict`	read
Draft outreach in-voice	`voice_draft`, `voice_classify_reply`	read
Advance a deal	`deals_update_stage`, `deals_bulk_create`	write
Invoice + chase payment	`invoice_prepare`, `invoice_record_qbo`, `invoice_payment_reconcile`, `ar_status`	mixed
Redline a contract	`contracts_run_playbook`, `contracts_get_playbook`, `contracts_set_playbook`	mixed
Source new brands	`active_spenders`, `competitive_landscape`, `brand_spend`, `sourcing_filter_new`	read
Sync a deal to your CRM / Airtable / Monday	`crm_import`, `stack_link_deal`, `stack_resolve_deal`, `stack_deal_refs`	mixed

Setup

1. Connect the Prscnt MCP

Add an HTTP MCP server in your client (Streamable HTTP, JSON responses):

{"mcpServers":{"prscnt":{"type":"http",
  "url":"https://mcp.prscnt.com/v1/mcp",
  "headers":{"Authorization":"Bearer <token>"}}}}

The bearer is either a Clerk OAuth token (claude.ai Add custom connector → Authorize) or a static prscnt_live_ / sk_live_ API key minted from the seat path once you're connected. First Authorize auto-provisions a free workspace — connect here.

2. Connect your own stack MCPs

Airtable and Monday are first-class bring-your-stack connectors alongside HubSpot, Asana, Notion, and DocuSign — plus Gmail, Drive, Calendar, QuickBooks, and Apollo as needed. Prscnt's check_integrations / connect_qbo / connect_apollo give guidance only; they store no credentials.

3. Run the link-then-mirror loop

crm_import pulls your brand-contact graph into Prscnt read-only → stack_link_deal points each Prscnt deal at its record in your system (id + URL only) → stack_resolve_deal for the reverse lookup. Both systems point at each other; Prscnt never copies your data.

Guardrails

Read vs. write. Intel tools are safe for an automated agent. Mutating tools (deals_update_stage, deals_bulk_create, the invoice_* writers, roster_* writers, voice_index_*) change state — gate them behind a human approval step in your portal.
Human approval on outbound. Prscnt drafts, never sends. Email, contracts, and invoices go through your own connector and a manager's sign-off.
Workspace is resolved from auth — never pass it. The server injects your workspace from the bearer; a mismatched id is rejected. Never send a workspace id from the client.
Billing and rate limits are real and shared. All access — interactive or headless — consumes one workspace's tier: token-bucket rate limits (HTTP 429 over cap), per-tool monthly allowances, and seat caps. Budget your agent's call volume.
Prscnt stores no third-party data or credentials. It holds pointers and its own intelligence. Your QuickBooks, Gmail, Airtable, and HubSpot data stay in your accounts.

Questions or want a hand wiring your portal: support@prscnt.com. See also the general docs.