Remove everything. Run anything.
One headless MCP layer for every SaaS — AI is the front end.
MCPaaS · Headless MCP Server
Every SaaS runs through one headless layer
Remove everything. Run anything.
One headless MCP layer for every SaaS — AI is the front end.
Your agents already write the interface and assemble the view. hdls is the layer underneath it: 37 SaaS backends on one substrate, each driven over MCP, so an agent can reach all of your data through a single connection.
Claude, GPT, or your own agent renders whatever view the question needs, on demand. The front end is yours, generated, then thrown away.
Choose what you need from 37 MCP-native products. Each is a live endpoint on one plane, metered per call, no seats.
On one entity graph they share data across each other, and with the teammates and partners you grant scoped keys. The join is native.
Each is a live MCP server on the same substrate, a usage-priced drop-in for the SaaS it replaces. Point your agent at the subdomain, hand it a scoped key, and it can work.
Name the product you want, sign in once, and connect it. No install, no keys to copy, no console to learn.
Point your agent at an endpoint like crm.hdls.ai. Swap it for any of the 37.
One account covers every product. No separate login per tool, nothing to wire up.
Approve once and your agent is working. Add billing or support later the same way.
Most stacks keep a customer in five disconnected systems. hdls keeps them as one entity, so a question that spans every tool is a single query against live data, not a data project.
ask "which enterprise accounts are at churn risk this week?" hdls resolves one entity across crm.hdls.ai plan = enterprise 342 accounts billing.hdls.ai + invoice status 18 overdue support.hdls.ai + open P1 (7d) 11 accounts product.hdls.ai + usage trend (30d) 23 declining → 6 accounts match every signal read 1,204 rows · 1 combined query · billed $0.019 (example) then draft a renewal email per account using crm + billing context, log the outreach to crm, open a save-play ticket in support 3 tools written · 1 governed action
No Zapier, no reverse-ETL, no warehouse to keep in sync. Tools share one substrate, so combining is a query instead of a pipeline.
The entity graph resolves a contact, a payer, and a sender into the same customer. Joins are semantic, not fuzzy string matching.
Every read spans tools against current data. No nightly sync, no stale copy, no lag between what happened and what the AI sees.
Combining is not read only. The AI finds across tools, then drafts, logs, schedules, and files back across tools in a single action.
Once the data is combinable, the real question is who can touch what. Every key, whether it is held by your own agent, a teammate, a contractor, or another company, is scoped down to the field and the row.
key hdls_live_8f2c…a91 held by acme-partners (external) expires 2026-09-01 scopes crm.hdls.ai read:contacts write:activity billing.hdls.ai read:invoices support.hdls.ai read:tickets write:tickets contracts.hdls.ai write:drafts sign: denied rows account.owner = "acme-partners" only audit on · every call attributed to holder
There's no app to learn. Each area below is a set of named MCP tools your agent calls directly — discover, install, connect, and do the work. Every block links to the docs.
Connect Claude, Cursor, Codex, or your own agent. No keys to paste, no SDK, no tenant ID to pass — your credential is pinned to one workspace.
/api/mcp to discover and install/api/mcp/<slug> for the actual workhdls_… API keys for headless server-to-server jobsCRM, Support, Billing, Inventory and more — each a schema with purpose-built, self-describing tools. Tailor any of them without touching the schema.
list_products & describe_product map the whole surfaceinstall_product flips a product on for your workspacecreate_account, move_deal_stageadd_custom_field for your own data points, no migrationAsk your AI to publish and hdls hosts a real page at a stable URL — no front-end to build, no server to run. Sandboxed, tenant-isolated, shareable.
publish_page for documents, dashboards, or intake formslist_pages & unpublish_page to manage and revokeFire on a database event or a schedule, gate it with conditions, then enqueue a task, post a signed webhook, send an email — or have hdls run an AI agent on the change for you, on OpenAI or Claude. No glue code.
create_trigger composes when, if, and do in one callinsert/update/delete) or schedule (every: "1h")run_agentReusable playbooks, persistent workspace context across sessions, and teammates who share the same backend — each scoped to a role, with a human in the loop on invites.
search_skills & get_skill pull in ready-made playbooksset_bio / set_preferenceinvite_teammate with reader / member / admin rolesIsolation is a database guarantee, not application code that has to remember to filter. Every claim is verifiable from your assistant.
X-Hdls-Signature HMAC-SHA256No dashboard, no forms to fill. You ask in plain language; your assistant calls the CRM's named MCP tools — create_account, log_activity, search_deals — and reads the answer straight back.
Add Acme as a customer, log a call, and show me this month's open deals.
On it — three named tool calls against crm.hdls.ai:
Done. Acme is in the CRM, the call is on its timeline, and you have 3 open deals worth $74k this month. Want me to draft follow-ups?
Illustrative session — real tool names, no fabricated capabilities. Swap any product the same way: connect support.hdls.ai, billing.hdls.ai, and the rest.
No marketing fog. Here is exactly what hdls is, how your data is protected, and what it costs.
Unlimited users, workspaces, and agents on every plan — each tier is metered by a monthly tool-call budget (the calls your agents make), enforced on every request. Start free.
Unlimited seats on every plan — pay for what your agents do, not per person. Annual saves ~17%. Each tier is enforced today by a per-minute rate limit and the monthly tool-call budget shown above; Custom is fully negotiated. *Record and storage volumes are guidance, not yet metered — finer-grained per-dimension caps are on the roadmap.
Get a scoped key, connect one endpoint, and let your agents read, combine, and act across the whole stack. Bring your own front-end.