SAGARIS

Remote MCP server for the SAGARIS AI CRM: read CRM contacts, calls, sequences and Brain views from an AI client. The few write tools each need their own API key scope.

Hosted MCP Server

npx add-mcp 'https://app.sagaris.ai/api/v1/mcp'

Installs into Claude Code, Codex, Cursor and more

Documentation

Connection

Model Context Protocol

Endpoint

https://app.sagaris.ai/api/v1/mcp POST only

Transport

Streamable HTTP, stateless JSON

Auth

Bearer token: your workspace API key, every message including the handshake

Scopes

7 of the platform's 12; OAuth sign-in grants 2

One JSON-RPC message per POST, one JSON response back, no session.

Your assistant reads the real record

Model Context Protocol is an open standard: it hands an AI client a set of tools it can call. SAGARIS runs the server half at /api/v1/mcp, over Streamable HTTP in stateless JSON mode: one JSON-RPC message per POST, one JSON response back, and no session to open or close.

Ask the assistant your team already works in about a deal and the answer comes from the record itself, not from a stale export pasted into a chat window. The endpoint is live and authenticated by workspace API key.

Safe to connect on day one

  • Reads the recordCRM, sequences and Brain tools read the record; they cannot edit contacts, dial or enroll people.
  • Drafts for approvalSocial checklist tools submit drafts and research findings for a person to approve. They never publish or approve. Social is coming soon.
  • Acts only through your agentLinkedIn execution is a separate, explicit permission, not something a read-only key inherits.
  • Labels itselfRead tools identify themselves as read only. Writes are marked as writes; the LinkedIn send carries destructive and open-world hints because an external message cannot be taken back, and its result report is marked destructive because it is final.

You decide what each key can see

It uses the workspace API key you already create in settings, sent as a bearer token. All of it is enforced in code:

  1. The key names the workspace. A client never gets to pick one.
  2. Scopes filter the tool list before the client sees it.
  3. Every message needs a valid key, the opening handshake included.
  4. 120 requests a minute per key; a browser client has its origin checked.

Close-up of a hand bringing a patch cable's plug to a brass-rimmed jack on a dark panel, more cabled jacks blurred behind it. No labels, no screen.

What your assistant can ask for.

The tools the endpoint offers today, with the scope each one needs. A key only ever sees the tools you granted it.

OAuth sign-in grants only linkedin.read and linkedin.execute.

ToolMarks itself
.read.read.read.execute.read.submit.record
Read the record10 tools
01sagaris_list_contacts requires contacts.readread
02sagaris_get_contact requires contacts.readread
03sagaris_list_sequences requires sequences.readread
04sagaris_get_linkedin_actions requires linkedin.readread
05sagaris_brain_claims_current requires contacts.readread
06sagaris_brain_contact_dossier requires contacts.readread
07sagaris_brain_nba_features requires contacts.readread
Social checklist3 toolsSocial: coming soonThe tools exist and submit drafts for a person to approve; the Social module is not shipped and no date is published.
08sagaris_list_social_checklist_items requires social_checklist.read Social: coming soonread
09sagaris_list_due_social_checklist_items requires social_checklist.read Social: coming soonread
10sagaris_get_social_checklist_item_context requires social_checklist.read Social: coming soonread
Draft for approval2 toolsSocial: coming soon
11sagaris_submit_social_checklist_item_output requires social_checklist.submit Social: coming soonwrite
12sagaris_record_social_checklist_item_evidence requires social_checklist.record Social: coming soonwrite
Act through your agent13 tools
13sagaris_get_profile requires linkedin.readread
14sagaris_linkedin_setup requires linkedin.executeread
15sagaris_linkedin_brief requires linkedin.executeread
16sagaris_linkedin_heartbeat requires linkedin.executewrite
17sagaris_linkedin_health_observation requires linkedin.executewrite
18sagaris_linkedin_claim requires linkedin.executewrite
19sagaris_linkedin_start requires linkedin.execute Destructive and open-world hints: an external message cannot be taken back.execute
Authorizes one invitation or message, with the exact recipient and copy, immediately before the agent performs it in its own LinkedIn session. Requires explicitly activated execution, a person's approval or an account set to run automatically, current sender health, and passing safety and quota checks.
20sagaris_linkedin_result requires linkedin.execute Destructive hint: a report is final and a sent one advances the sequence.write
21sagaris_linkedin_sync requires linkedin.executewrite
22sagaris_linkedin_connection_accepted requires linkedin.executewrite
23sagaris_linkedin_schedule_report requires linkedin.executewrite
24sagaris_linkedin_run_report requires linkedin.executewrite
25sagaris_linkedin_reconnect_ack requires linkedin.executewrite

CRM and Brain tools read; Social checklist tools write drafts and research findings, never publish or approve. LinkedIn execution is separate: an explicitly enabled external agent can act only where a person approved the action or set its account to run automatically, subject to safety and quota checks. It acts only in the campaigns it was granted or, when none were named, in every campaign on its LinkedIn account; every agent connected through OAuth has none named. A read-only key cannot execute actions.

You choose what leaves the workspace

The Brain holds more kinds of record than this endpoint hands out, and every one of those calls was made on purpose.

Stays inside the workspace

  • Internal delivery commitmentsThe record of what your own team promised a customer.Not served
  • RequirementsNot served
  • Machine-derived competitor intelligenceIt only reads properly next to its source and its date.Not served

The list is compiled from the same table the runtime check uses; a new subject type stops the build until someone decides.

LinkedIn, three modes

  • manual defaultDefault for every account
  • review A person approved the action
  • auto The account is set to run automatically

A connected external agent can claim and start a LinkedIn action only when execution is explicitly activated, a person approved the action or set its account to run automatically, and sender, campaign, safety and quota checks pass.

The agent sends from its own LinkedIn session; SAGARIS does not host that session.

Revoke is the off switch

Revoke the API key and access ends immediately, because the key is the only credential the endpoint accepts.

No refresh token is issued, so no session outlives the key.

Included in your seat

$499

per founding seat per month, every module and every agent included.

Pre-launchSAGARIS is pre-launch. We are running a small number of founding customers now.

What the seat covers

No per-call charge and no higher tier gating the endpoint: it ships with the product rather than in a tier above it. Calls are rate limited per key; LinkedIn actions count against the plan's allowance.

The questions this endpoint gets asked.

Short answers, including the ones about what it will not do.

More questions? Book a demo

  • What is the SAGARIS MCP endpoint?

    A Model Context Protocol server at /api/v1/mcp that lets an AI client read your SAGARIS workspace, work through your Social checklist (Social is coming soon), coordinate gated LinkedIn actions with a connected external agent, and subscribe that agent to three LinkedIn events delivered as signed webhooks. With MCP Events, a client that takes webhooks subscribes to linkedin.work.available, linkedin.schedule.changed and linkedin.reconnect.requested and is told when to act, in ids, counts and the account's label in SAGARIS, never a contact's name, a profile URL or message copy. Each read and write capability needs its own scope. It speaks Streamable HTTP in stateless JSON mode and authenticates with your workspace API key as a bearer token.
  • Can an assistant change anything in my workspace through it?

  • What can the SAGARIS MCP server read?

  • How does it know which workspace to read?

  • Which MCP clients can connect to SAGARIS?

  • What does it cost?

  • How do I turn off MCP access to my SAGARIS workspace?

It is already included in your seat.

Once your workspace is set up, create an API key in settings and point your client at it. The platform page shows everything it can reach.

Developers and integrations