SymbiozAI

MCP Server

The MCP-only CRM.

Headless by design. Operated by your AI agent via MCP. Claude Code, Cursor, GPT. It's all native.

Install the MCP in 5 minTry the live sandboxBook a meeting

01 · QUICKSTART

Ship your first CRM call in 5 minutes.

One command. The CLI detects your client and configures the connection. No API key gymnastics. No JSON to edit by hand.

1

Step 1

Auto-detects Claude Code, Cursor, Cline, Goose, Continue.dev. Writes the config. Done.

bash
npx @symbiozai/mcp-setup
2

Step 2

Your agent now has 35 SymbiozAI MCP missions available. No plugin. No SDK. Standard MCP spec.

bash
"What missions does SymbiozAI expose?"
3

Step 3

Run your opening mission in natural language. Your agent calls start_targeting. Pipeline populated. Under 60 seconds.

bash
"Target 50 Series A SaaS founders in France using Claude Code."

02 · AGENT CONFIGS

Drop-in config for your agent.

SymbiozAI implements stdio and HTTP+SSE transports. Copy-paste the snippet for your client. That is all.

Claude Code

claude_desktop_config.json
{
  "mcpServers": {
    "symbiozai": {
      "command": "npx",
      "args": ["@symbiozai/mcp-setup", "--mode", "stdio"],
      "env": { "SYMBIOZAI_TOKEN": "your-token" }
    }
  }
}

Add this to your claude_desktop_config.json. Restart Claude Code. SymbiozAI appears in your tool list.

Cursor

~/.cursor/mcp.json
{
  "symbiozai": {
    "url": "https://api.symbioz.ai/mcp/sse",
    "headers": { "Authorization": "Bearer your-token" }
  }
}

HTTP+SSE transport. Token from your SymbiozAI dashboard. Cursor picks it up on next launch.

Cline · Goose · Continue.dev

shell
# Auto-detects the client and writes the correct config format
npx @symbiozai/mcp-setup

The installer reads your environment, detects which client you run, and writes the correct config format automatically.

03 · MISSION CATALOG

35 missions. Full sales cycle.

Every mission is designed to be called by a language model in natural language. Each has a schema, an example prompt, and an expected output. No guesswork.

Acquisition

7 missions
  • start_targeting

    Target ICP-fit prospects from 23 data providers. Company + contact.

    Green
  • enrich_contact

    Enrich a contact with email, phone, LinkedIn, social data.

    Green
  • enrich_company

    Enrich a company with firmographic data, funding, headcount, tech stack.

    Green
  • get_buying_signals

    Detect intent signals: job postings, funding rounds, tech changes.

    Green
  • reactivate_dormant_leads

    Surface dormant leads with re-engagement signals.

    Orange
  • search_contacts

    Semantic search across your contact database.

    Green
  • get_lead_context

    Full 360 view of a lead: history, interactions, enrichment.

    Green

Qualification

6 missions
  • qualify_lead

    Composite qualification with structured reasoning. Replaces manual MQL/SQL scoring.

    Green
  • analyze_communication_style

    DISC-based profiling from LinkedIn, emails, meeting history.

    Green
  • get_meeting_prep_brief

    Pre-call brief: company context, contact history, recommended talking points.

    Green
  • get_pipeline_snapshot

    Real-time pipeline health: by stage, by rep, by deal value.

    Green
  • assess_deal_health

    0-100 deal score. Every factor explained. No blackbox.

    Green
  • poll_events

    Monitor pipeline for hot signals: replies, stage moves, silence.

    Green

Engagement

8 missions
  • draft_email_personalized

    DISC-aware email draft. Adapts tone to contact profile.

    Orange
  • draft_followup_sequence

    Multi-step follow-up sequence, DISC-aware, throttle-controlled.

    Orange
  • summarize_meeting_recording

    Summarize a meeting transcript and extract action items.

    Green
  • update_deal

    Update a deal: stage, amount, close date, owner.

    Orange
  • book_meeting

    Create a meeting in calendar, link to deal.

    Orange
  • send_email

    Send email via connected mailbox (HITL: Orange class by default).

    Red
  • create_contact

    Create a new contact with full enrichment.

    Orange
  • create_deal

    Create a new deal, linked to contact and company.

    Orange

Meta / Control

7 missions
  • escalate_to_human

    Surface a decision to the human supervisor with context.

    Green
  • get_audit_log

    Retrieve action history for a contact, deal, or tenant.

    Green
  • get_supervision_queue

    List of Orange/Red actions pending human review.

    Green
  • run_playbook

    Execute a pre-defined playbook (sequence of missions).

    Varies
  • get_agent_status

    Current agent state, running missions, pending actions.

    Green
  • export_my_data

    GDPR art. 15 native. Export all your data on demand.

    Green
  • kill_switch

    Immediately halt all agent activity for the tenant.

    Red

Phase 2 - 7 missions coming: score_company, map_stakeholders, detect_competitor_displacement, analyze_win_loss, forecast_pipeline, create_sequence, push_to_outbound.

04 · ARCHITECTURE

23 providers. One endpoint.

SymbiozAI does not rebuild data sources. It wraps them. Your agent calls one mission. SymbiozAI picks the right provider, aggregates, deduplicates, and returns a structured response.

DATA SOURCES

  • enrich_contactContact enrichment
    ApolloBrightDataHunterClearbitLusha
  • enrich_companyCompany enrichment
    CrunchbasePappersINSEEClearbitBrightData
  • intentIntent signals
    6senseBomboraFunding data
  • commsCommunication
    Unipile
  • socialSocial
    LinkedIn
  • registryRegistry
    Pappers (FR)INSEE (FR)Crunchbase

MCP ENDPOINT

mcp

SymbiozAI MCP server

api.symbioz.ai/mcp

endpoint
1
enrich. missions
2
providers
23

> enrich_contact · enrich_company

23 providers. 2 enrichment missions. No API key per provider. No aggregation code to maintain.

05 · HUMAN-IN-THE-LOOP

Every action has a class. Every class has a behavior.

The HITL policy is policy-as-code. Every mission is assigned a class at deploy time. You configure thresholds. The system enforces them. No ad-hoc overrides.

Green

Behavior
Runs automatically. No confirmation needed. The agent reads, enriches, qualifies, and scores without touching your queue.
Examples
enrich_contact, qualify_lead, assess_deal_health, get_pipeline_snapshot, analyze_communication_style

Orange

Behavior
Agent proposes. You confirm. The action runs as dry-run first. You see the proposed output in your queue. One click to approve or reject. Nothing ships until you say so.
Examples
draft_email_personalized, update_deal, book_meeting, create_contact

Red

Behavior
Blocked until you approve. The action does not execute until you give explicit approval. No queuing, no it-ran-while-you-slept.
Examples
send_email, kill_switch, bulk sequence launch

Policy-as-code

json
{
  "mission": "draft_email_personalized",
  "hitl_class": "orange",
  "auto_approve_threshold": null,
  "requires_human_id": true
}

Default policy is conservative. Tune per mission category from your tenant config. Every change is logged. AI Act article 14 compliance is structural, not a checkbox. The HITL classes are the article-14 implementation.

06 · DIFFERENTIATORS

Three missions no other MCP CRM exposes.

These are not integrations. They are proprietary missions built on top of your contact and deal data, running inside SymbiozAI's reasoning layer.

analyze_communication_style

DISC profiling

Your agent reads available data: LinkedIn activity, email history, meeting transcripts. It produces a DISC profile with writing guidance. You write to a Dominant differently than to an Influential. The reasoning is exposed, not hidden in a model.

Example prompt

"Analyze Thomas Martin's communication style before my call tomorrow."

Expected output

DISC profile (D/I/S/C scores), preferred communication style, email tone recommendations, topics to avoid.

assess_deal_health

Multi-signal score

Every deal scored 0-100. Five signal sources: email reply rate, meeting cadence, deal velocity vs stage average, stakeholder engagement breadth, competitive signals. Every factor documented. No blackbox.

Example prompt

"Why is the Acme deal at risk? What should I do this week?"

Expected output

Score 0-100, factor breakdown, recommended next actions, urgency classification.

qualify_lead

Composite gatekeeper

Structured qualification against your ICP criteria with explicit reasoning. Consistent across every rep. Repeatable, auditable, no MQL/SQL committee.

Example prompt

"Is Sophie Durand from TechVision worth pursuing for our Series A ICP?"

Expected output

Qualified / not qualified, ICP match score, blocking factors, recommended next step.

07 · SUPERVISION

You don't operate the CRM. You supervise sensitive decisions.

5 minutes a day. A queue of Orange and Red actions. You approve, reject, or ask the agent to revise. Everything else runs automatically.

What's in the queue

Every Orange action lands here before execution. Every Red action is blocked here until you act. The queue is sorted by urgency and mission type.

Action detail

For each queued action: full context, the agent's reasoning, the proposed output, and the impact if approved or rejected. You have what you need to decide.

One click

Approve. Reject. Ask the agent to revise with a note. The agent acts on your decision within seconds.

Audit trail

Every human decision is logged: your identity, the timestamp, your reasoning if you left a note. Immutable. Part of your AI Act article 14 compliance record.

The console is not the product. The MCP server is. The console is the control room for the 5 decisions a day that require you.

08 · COMPLIANCE

EU-hosted. AI Act native. GDPR built-in.

Four facts. All verifiable. No marketing claims.

EU-hosted

Infrastructure in Frankfurt (DigitalOcean FRA1). Your data does not leave the EU. No exceptions.

AI Act art. 14

The HITL 3-class policy is the article-14 implementation. Every sensitive action goes through a human gate. Immutable audit log. 7-year retention. Kill-switch propagation under 1 second.

GDPR art. 15

GET /audit/my-data - export all your data on demand. No locked-in data. No support ticket to file.

LLM-agnostic

UnifiedLLMClient multi-provider. No fine-tuning on your data. No data retention by LLM providers. You can swap the underlying model without migrating your data.

09 · AUDIT LOG

Every action logged. Nothing editable. Nothing deletable.

HMAC-SHA256 chained entries. 7-year retention. Break-glass access via audit_admin. Every MCP call, every human decision, every kill-switch event - in the log.

Append-only

No record can be modified after creation. No delete endpoint exists. Audit integrity is structural.

HMAC-chained

Each log entry contains the HMAC-SHA256 hash of the previous entry. Tampering with any entry breaks the chain. Verification is programmatic.

7-year retention

Compliant with AI Act article 14 and French commercial accounting law. Log rotation does not delete - it archives.

Queryable

GET /audit/my-data returns your full export in structured JSON. Filter by contact, deal, agent session, or date range.

Human-attributed

Every human approval, rejection, or note is stored with the authenticated user's identity and timestamp. You know who approved what, when, and why.

Export your audit log
# Export your audit log
curl -H "Authorization: Bearer $TOKEN" \
  https://api.symbioz.ai/audit/my-data

10 · PRICING

Free during beta. Usage-based after.

No seat pricing. No annual contract. No per-user lock-in. You pay for what your agent calls.

Beta - free now

500 MCP calls/day at no cost. No credit card required to start. The limit resets every 24 hours. Overage: €0.15/credit pay-per-use. Top up from your dashboard when needed.

Post-beta

Usage-based pricing. The model will be announced 30 days before beta closes. No retroactive changes to calls already made.

What counts as a call

One MCP mission call = one credit. A start_targeting that returns 50 prospects = 1 credit. An enrich_contact = 1 credit. LLM inference within a mission is included.

Questions about pricing for larger volumes: Book a meeting

11 · GET STARTED

Connect your agent. Now.

Five minutes. One command. 35 missions available.

Install the MCP in 5 minTry the live sandbox

Or request a live walkthrough: Book a meeting

Full docs at docs.symbioz.ai/mcp