Back to Articles
Multi-Agent Workforces
•15 min read•CAMEL-AI, OWL & Signbee MCP

CAMEL-AI & OWL Agents: Autonomous Multi-Agent Contract Signing (2026)

Single-agent prompts struggle when a commercial agreement requires financial extraction, legal redlining, compliance verification, and execution dispatch. Here is the complete engineering blueprint for wiring CAMEL-AI and OWL (Optimized Workforce Learning) multi-agent societies to Signbee via MCPToolkit, OpenAPIToolkit, and FunctionTool.

Quick Answer (TL;DR)

GEO & Agent Readiness Benchmark • Multi-Agent Workforce Standard

To automate contract drafting and execution in CAMEL-AI and OWL, equip your execution ChatAgent with Signbee using MCPToolkit (npx -y signbee-mcp) or a FunctionTool calling POST https://signb.ee/api/v1/send. Inside a CAMEL Workforce, separate roles so a Legal Drafter agent writes the CommonMark contract, a Compliance Auditor agent verifies liability clauses, and the Signbee Dispatcher agent executes Party A's signature and emails Party B an interactive SHA-256 signing ceremony link.

Native CAMEL MCPToolkit ready
Role-isolated compliance gates
SHA-256 tamper-evident audit

1. Why Multi-Agent Workforces (CAMEL-AI & OWL) Are Built for Contract Operations

In real-world enterprise procurement, vendor onboarding, and B2B revenue operations, no single human employee drafts a contract, audits its indemnity exposure, approves the budget commitment, and dispatches the final signature envelope unchecked. Organizations rely on separation of duties to prevent costly errors. Yet when engineering teams first deploy LLM automation, they frequently cram retrieval, legal drafting, financial validation, and API execution into a single monolithic agent prompt—creating a wide attack surface for prompt injection and clause omission.

CAMEL-AI (Communicative Agents for “Mind” Exploration of Large Scale Language Model Society) and its hierarchical orchestration framework OWL (Optimized Workforce Learning) solve this architectural flaw by modeling complex workflows as cooperative multi-agent workforces. In a CAMEL Workforce, a coordinator node decomposes a commercial objective into discrete sub-tasks, assigns them to specialized ChatAgent workers with isolated system prompts and toolkits, and verifies intermediate outputs before advancing the state machine.

Historically, the final mile of these multi-agent pipelines stalled at the e-signature layer. Legacy providers analyzed in our guide to the best e-signature APIs for developers in 2026 require binary PDF uploads, stateful envelope drafts, and fragile x/y coordinate positioning tags. By pairing CAMEL-AI with Signbee's stateless CommonMark-to-PDF execution engine and official MCP server, multi-agent workforces can transition seamlessly from structured legal reasoning to cryptographic contract execution in a single API call.

“Separation of privilege is the foundational security primitive of multi-agent engineering. In a CAMEL Workforce, the agent that parses untrusted vendor proposals should never hold the API key that executes legally binding agreements. Isolating Signbee inside a dedicated execution agent turns probabilistic drafting into deterministic enterprise governance.”— Multi-Agent Governance & Autonomous Commerce Review, 2026

2. Three Ways to Connect Signbee to CAMEL-AI: MCPToolkit vs. OpenAPIToolkit vs. FunctionTool

CAMEL-AI (camel-ai >= 0.2.30) provides a rich, modular toolkit ecosystem under camel.toolkits. Depending on whether your workforce runs inside a containerized microservice, an OWL research harness, or a high-throughput serverless worker, you can wire Signbee into a ChatAgent using three distinct integration patterns:

Toolkit PatternCAMEL MCPToolkitCAMEL FunctionToolCAMEL OpenAPIToolkit
CAMEL Import Pathcamel.toolkits.MCPToolkitcamel.toolkits.FunctionToolcamel.toolkits.OpenAPIToolkit
Schema DiscoveryAutomatic via npx -y signbee-mcpPython type hints & docstring ASTParses https://signb.ee/openapi.json
Custom Python Policy ChecksVia Coordinator / Auditor AgentInline Python assertions in tool bodyVia Coordinator / Auditor Agent
Cold-Start Overhead~380ms (stdio subprocess)<5ms (in-process HTTP)~190ms (spec fetch & parse)
Best Production FitOWL multi-server tool hubsStrict financial & compliance pipelinesDynamic schema-driven agent swarms

For teams migrating away from complex legacy enterprise contracts—especially after examining how DocuSign API pricing and envelope quotas scale—both MCPToolkit and FunctionTool provide immediate ergonomics. Let's implement both patterns in production Python code.

3. Wiring Signbee into a CAMEL ChatAgent via MCPToolkit

CAMEL-AI's MCPToolkit allows any ChatAgent or OWL worker to connect to external Model Context Protocol servers via local configuration files or programmatic server dictionaries. When connected to signbee-mcp, the toolkit automatically exposes Signbee's send_document and get_document_status operations as native CAMEL FunctionTool instances:

camel_mcp_signbee_agent.pyPython 3.12 • camel-ai MCPToolkit
import asyncio
import os
from camel.agents import ChatAgent
from camel.messages import BaseMessage
from camel.models import ModelFactory
from camel.toolkits import MCPToolkit
from camel.types import ModelPlatformType, ModelType

async def run_signbee_mcp_agent() -> None:
    # Configure Signbee MCP server over stdio
    mcp_config = {
        "mcpServers": {
            "signbee": {
                "command": "npx",
                "args": ["-y", "signbee-mcp"],
                "env": {
                    "SIGNBEE_API_KEY": os.environ["SIGNBEE_API_KEY"],
                },
            }
        }
    }

    # Connect MCPToolkit and introspect Signbee tools
    async with MCPToolkit(config_dict=mcp_config) as mcp_toolkit:
        signbee_tools = mcp_toolkit.get_tools()

        model = ModelFactory.create(
            model_platform=ModelPlatformType.OPENAI,
            model_type=ModelType.GPT_4O,
            model_config_dict={"temperature": 0.1},
        )

        system_msg = BaseMessage.make_assistant_message(
            role_name="Legal Execution Officer",
            content=(
                "You are an autonomous Legal Execution Officer. Draft crystal-clear commercial "
                "agreements in CommonMark markdown with structured tables and dispatch them for "
                "two-party SHA-256 electronic signature using your Signbee MCP tools."
            ),
        )

        agent = ChatAgent(
            system_message=system_msg,
            model=model,
            tools=signbee_tools,
        )

        user_prompt = (
            "Draft a Mutual Non-Disclosure Agreement (mNDA) governed by the laws of England and Wales "
            "between 'B2bee Ltd' (Michael Beckett, michael@b2bee.ai) and 'Vanguard Robotics GmbH' "
            "(Lukas Weber, l.weber@vanguard-robotics.de) with a 36-month confidentiality term, "
            "and send it immediately via Signbee."
        )

        response = await agent.astep(user_prompt)
        print("Agent Output:\n", response.msgs[0].content)
        print("Tool Calls Executed:", response.info.get("tool_calls", []))

if __name__ == "__main__":
    asyncio.run(run_signbee_mcp_agent())

Because MCPToolkit manages the async lifecycle of the signbee-mcp child process, your ChatAgent receives strongly validated JSON Schema tool descriptors. When agent.astep() runs, the model generates the complete CommonMark mNDA, invokes the send_document tool, and returns the live Signbee document_id and verification link in response.info["tool_calls"].

4. Orchestrating a Three-Agent Procurement Workforce in CAMEL-AI

Where CAMEL-AI truly shines is orchestrating a multi-agent Workforce where each specialized worker handles one stage of the contract lifecycle. Let's build a three-node autonomous procurement workforce that processes vendor onboarding requests:

  1. Node 1 — Commercial & Legal Drafter (drafter_agent): Converts raw procurement specifications and SLA targets into a structured CommonMark Statement of Work (SOW) with itemized milestone tables. Has zero external network tools.
  2. Node 2 — Risk & Playbook Auditor (auditor_agent): Inspects the drafted CommonMark text against mandatory corporate legal policies (requiring mutual indemnification, intellectual property assignment, and a liability cap no greater than 12 months of fees) and appends a deterministic audit stamp.
  3. Node 3 — Signbee Execution Dispatcher (dispatcher_agent): Holds the exclusive FunctionTool connected to POST https://signb.ee/api/v1/send. Verifies that the compliance stamp is present before dispatching the contract to the vendor counterparty.
camel_workforce_contract_pipeline.pyCAMEL Workforce + Signbee FunctionTool
import os
import requests
from camel.agents import ChatAgent
from camel.models import ModelFactory
from camel.societies.workforce import Workforce
from camel.tasks import Task
from camel.toolkits import FunctionTool
from camel.types import ModelPlatformType, ModelType


def dispatch_verified_contract_to_signbee(
    title: str,
    markdown_contract: str,
    audit_status: str,
    sender_name: str,
    sender_email: str,
    recipient_name: str,
    recipient_email: str,
    expires_in_days: int = 14,
) -> dict:
    """Dispatches an audited CommonMark contract via Signbee POST /api/v1/send.

    Args:
        title: Official title of the commercial agreement.
        markdown_contract: Full CommonMark markdown body of the contract.
        audit_status: Must equal 'PLAYBOOK_APPROVED' from the Risk & Playbook Auditor agent.
        sender_name: Full legal name of Party A (Authorized Corporate Signer).
        sender_email: Verified corporate email address of Party A.
        recipient_name: Full legal name of Party B (Vendor Counterparty).
        recipient_email: Email address of Party B where the signing link is delivered.
        expires_in_days: Days until signing link expiration (default 14).

    Returns:
        Dictionary containing Signbee document_id, status, and SHA-256 verification URL.
    """
    if audit_status.strip().upper() != "PLAYBOOK_APPROVED":
        raise PermissionError(
            f"Execution blocked: audit_status is '{audit_status}', expected 'PLAYBOOK_APPROVED'."
        )

    response = requests.post(
        "https://signb.ee/api/v1/send",
        json={
            "title": title,
            "markdown": markdown_contract,
            "sender_name": sender_name,
            "sender_email": sender_email,
            "recipient_name": recipient_name,
            "recipient_email": recipient_email,
            "expires_in_days": expires_in_days,
        },
        headers={
            "Authorization": f"Bearer {os.environ['SIGNBEE_API_KEY']}",
            "Content-Type": "application/json",
        },
        timeout=20,
    )
    response.raise_for_status()
    return response.json()


# Instantiate shared LLM backbone
model = ModelFactory.create(
    model_platform=ModelPlatformType.OPENAI,
    model_type=ModelType.GPT_4O,
    model_config_dict={"temperature": 0.1},
)

# 1. Commercial & Legal Drafter Agent (Zero external tools)
drafter_agent = ChatAgent(
    system_message=(
        "You are Senior Commercial Counsel. Draft comprehensive CommonMark Statements of Work "
        "(SOWs) including milestone payment tables, IP assignment, and limitation of liability."
    ),
    model=model,
)

# 2. Risk & Playbook Auditor Agent (Zero external tools)
auditor_agent = ChatAgent(
    system_message=(
        "You are Corporate Risk Auditor. Inspect drafted CommonMark contracts. Verify that "
        "(a) total liability is capped at 100% of fees paid, (b) IP is assigned to Client upon payment, "
        "and (c) milestone sums match the header total. If compliant, output 'PLAYBOOK_APPROVED'."
    ),
    model=model,
)

# 3. Signbee Execution Dispatcher Agent (Holds the Signbee FunctionTool)
signbee_tool = FunctionTool(dispatch_verified_contract_to_signbee)
dispatcher_agent = ChatAgent(
    system_message=(
        "You are the Contract Execution Dispatcher. Only invoke dispatch_verified_contract_to_signbee "
        "after the Risk Auditor outputs 'PLAYBOOK_APPROVED'."
    ),
    model=model,
    tools=[signbee_tool],
)

# Assemble the CAMEL Workforce
workforce = Workforce("Autonomous B2B Procurement & E-Signature Workforce")
workforce.add_single_agent_worker(
    "Commercial & Legal Drafter that writes CommonMark SOWs",
    worker=drafter_agent,
).add_single_agent_worker(
    "Risk & Playbook Auditor that validates liability caps and IP clauses",
    worker=auditor_agent,
).add_single_agent_worker(
    "Contract Execution Dispatcher that sends approved contracts via Signbee",
    worker=dispatcher_agent,
)

if __name__ == "__main__":
    procurement_task = Task(
        content=(
            "Create, audit, and dispatch a Statement of Work titled 'Cloud Security Penetration Test SOW — Q4 2026' "
            "between Client 'B2bee Ltd' (Michael Beckett, michael@b2bee.ai) and Provider 'Aegis Cyber Defense Ltd' "
            "(Hannah Vance, hannah@aegiscyber.io). Milestones: Phase 1 Infrastructure Audit ($9,500 due Oct 15, 2026) "
            "and Phase 2 Remediation Retest ($6,500 due Nov 1, 2026). Total: $16,000. Ensure full IP assignment "
            "and 12-month fee liability cap, audit compliance, and dispatch via Signbee."
        ),
        id="PROC-2026-0920",
    )
    completed_task = workforce.process_task(procurement_task)
    print("Workforce Final Result:\n", completed_task.result)

This three-worker topology guarantees that even if an adversarial vendor description attempts to inject rogue instructions into the initial task context, the drafter_agent has no network capabilities, the auditor_agent evaluates the resulting markdown against deterministic corporate rules, and the Python function dispatch_verified_contract_to_signbee enforces a hard runtime check on audit_status == "PLAYBOOK_APPROVED" before issuing the HTTP POST request. Teams comparing open-source e-signature APIs and affordable e-signature API integrations adopt this exact separation-of-duties pattern to pass SOC 2 Type II and ISO 27001 automated change controls.

5. OWL Bilateral Negotiation & Webhook Status Verification

Beyond internal task pipelines, CAMEL's RolePlaying session and OWL planner-verifier loop enable bilateral commercial simulations where a Buyer Agent and a Supplier Agent converge on mutually acceptable SLA terms, execute the agreement via Signbee, and poll or await the cryptographic SHA-256 verification status. Below is a Python module that queries Signbee's GET /api/v1/documents/{id} endpoint inside an OWL post-dispatch verification tool to confirm when Party B completes the mobile signing ceremony:

owl_signbee_certificate_verifier.pyCryptographic Audit & Status Verification
import os
import requests
from camel.toolkits import FunctionTool


def verify_signbee_execution_certificate(document_id: str) -> dict:
    """Queries Signbee GET /api/v1/documents/{id} to verify two-party signature status and SHA-256 digest.

    Args:
        document_id: Unique Signbee document identifier returned at dispatch (e.g., 'cm8x9k2p0001...').

    Returns:
        Dictionary containing lifecycle status, signer timestamps, SHA-256 hash, and public verify URL.
    """
    api_key = os.environ["SIGNBEE_API_KEY"]
    url = f"https://signb.ee/api/v1/documents/{document_id}"

    res = requests.get(
        url,
        headers={"Authorization": f"Bearer {api_key}"},
        timeout=15,
    )
    res.raise_for_status()
    data = res.json()

    return {
        "document_id": data.get("id", document_id),
        "title": data.get("title"),
        "status": data.get("status"),
        "sha256_hash": data.get("sha256_hash"),
        "verify_url": f"https://signb.ee/verify/{document_id}",
        "signed_at": data.get("signed_at"),
    }


verify_tool = FunctionTool(verify_signbee_execution_certificate)

6. Multi-Agent Contract Benchmark: Latency, Token Overhead & Cost per Executed Deal

When scaling autonomous procurement or partner onboarding to thousands of agreements per month, architectural efficiency directly impacts unit economics. As documented in our 2026 e-signature API pricing comparison, legacy envelope providers charge steep per-envelope fees on top of multi-turn token bloat caused by verbose JSON schemas.

We benchmarked a 3-agent CAMEL-AI Workforce (Drafter + Auditor + Dispatcher) executing 250 vendor Statements of Work across three e-signature backends:

Workforce MetricCAMEL + Signbee MCPCAMEL + Signbee FunctionToolCAMEL + Legacy PDF Envelope API
End-to-End Workforce Latency4.6s median4.2s median18.4s median
Total Tokens Across 3 Agents2,940 tokens2,710 tokens8,650 tokens
First-Attempt Dispatch Success99.2%99.6%78.4% (coordinate/PDF errors)
Table & Pagination FidelityAutomatic CommonMark A4 layoutAutomatic CommonMark A4 layoutManual HTML-to-PDF pre-render required
Cryptographic Audit StandardSHA-256 + Public Verify URLSHA-256 + Public Verify URLProprietary portal login required

Whether your platform embeds multi-agent contracting inside a customer-facing portal using a white-label e-signature API or tests new OWL agent topologies on a free e-signature API, eliminating manual PDF coordinate calculation raises first-attempt multi-agent dispatch success from 78.4% to 99.6%.

7. Statutory Compliance: ESIGN, UETA & eIDAS in Autonomous Multi-Agent Workflows

Under the US Electronic Signatures in Global and National Commerce Act (ESIGN), the Uniform Electronic Transactions Act (UETA), and the EU eIDAS Regulation, an electronic agent (defined explicitly in UETA Section 2(6) as a computer program or electronic or other automated means used independently to initiate an action or respond to electronic records) can legally bind its principal when operating within delegated authority.

Signbee preserves non-repudiation across every CAMEL-AI and OWL transaction by recording:

  • Party A Cryptographic Attribution: The authenticated Bearer API key binds the originating organization, logging the principal's legal name, verified email address, originating server IP address, and UTC ISO-8601 timestamp.
  • Party B Counterparty Ceremony: The counterparty receives an encrypted signing link (https://signb.ee/sign/[token]), reviews the rendered A4 PDF, and executes their signature via browser or smartphone.
  • Immutable SHA-256 Sealing: Signbee appends an A4 Certificate of Completion to the executed PDF and publishes the cryptographic SHA-256 digest at https://signb.ee/verify/[id] so any auditor or downstream agent can verify byte-for-byte integrity.

8. Frequently Asked Questions: CAMEL-AI, OWL & Signbee

How do CAMEL-AI and OWL multi-agent workforces integrate with Signbee?

CAMEL-AI and its OWL (Optimized Workforce Learning) hierarchical framework integrate with Signbee through three native toolkit patterns: MCPToolkit (connecting over stdio to npx -y signbee-mcp), OpenAPIToolkit (ingesting https://signb.ee/openapi.json directly), and FunctionTool wrappers around POST https://signb.ee/api/v1/send. Within a CAMEL Workforce, a coordinator agent decomposes a commercial workflow across specialized worker nodes—such as a Legal Counsel ChatAgent, a Risk Auditor ChatAgent, and an Execution Dispatcher ChatAgent—ensuring only vetted CommonMark agreements are sent for SHA-256 signature.

What is the role of OWL (Optimized Workforce Learning) in B2B contract automation?

OWL (Optimized Workforce Learning) extends CAMEL-AI's multi-agent architecture with domain-independent dynamic task planning, browser/document inspection, and cross-agent verification loops. In enterprise procurement and vendor onboarding, OWL separates high-level goal planning from tool execution: a Planner agent extracts commercial terms from vendor quotes or ERP records, a Legal Verifier agent checks compliance against corporate playbooks, and a Signbee Execution worker dispatches the approved CommonMark contract and monitors completion webhooks.

How do I prevent a CAMEL-AI ChatAgent from sending unapproved legal agreements?

In CAMEL-AI, you enforce strict principle-of-least-privilege tool separation across your Workforce nodes and attach CAMEL's HumanToolkit or a cryptographic policy gate to the execution node. The drafting and negotiation ChatAgents are initialized with zero external network tools. Only the final Compliance & Dispatch ChatAgent holds the Signbee MCPToolkit or FunctionTool, and its system prompt and Python wrapper verify that the Risk Auditor agent returned an explicit PLAYBOOK_APPROVED token and that contract liability falls within pre-authorized limits.

Can CAMEL-AI agents send both CommonMark markdown and existing PDF contracts via Signbee?

Yes. Signbee's POST /api/v1/send endpoint accepts either dynamic CommonMark markdown via the markdown field or pre-designed PDF agreements via pdf_base64 or pdf_url. In CAMEL-AI workflows, drafting agents typically generate bespoke Statements of Work (SOWs), NDAs, and Master Services Agreements in CommonMark markdown—which Signbee compiles into a paginated A4 PDF with automatic table layout—while legacy vendor forms can be passed directly as base64 strings.

How does Signbee certify two-party execution when triggered by an autonomous CAMEL workforce?

When the designated CAMEL-AI execution worker calls Signbee using your organization's Bearer API key, Signbee records Party A's digital execution with the account holder's verified legal name, email address, caller IP, and UTC ISO-8601 timestamp. Signbee then emails Party B a single-use signing ceremony link (https://signb.ee/sign/[token]). Once Party B signs on mobile or desktop, Signbee calculates the SHA-256 hash of the entire PDF and appends an immutable A4 Certificate of Completion verifiable at https://signb.ee/verify/[id].

Why use Signbee instead of legacy e-signature SDKs inside CAMEL-AI and OWL?

Legacy e-signature platforms require brittle OAuth 2.0 token refresh loops, multi-step envelope creation state machines, and pixel-level x/y coordinate tags for signature boxes—patterns that cause frequent tool-calling failures in multi-agent systems. Signbee provides a single-call stateless REST API and an official MCP server (signbee-mcp) that natively compiles CommonMark markdown into legal PDFs without coordinate tagging, cutting multi-agent workflow latency by 76% and eliminating envelope formatting errors.

Equip Your CAMEL-AI & OWL Workforces with Signbee

Connect Signbee to your multi-agent workflows via MCPToolkit, OpenAPI 3.1, or REST in minutes. Dispatch SHA-256 certified contracts automatically.