Agent Architecture & Orchestration

AI Agents on BotDigit MCP

BotDigit transforms static project repositories into collaborative environments where human engineers and autonomous AI agents work alongside each other. Through MCP, agents gain typed access to plan sprints, update tasks, link evidence, and prepare approvals.

Core Agent Archetypes

Whether deploying custom LangChain agents, Claude Desktop workflows, or AutoGPT workers, agents on BotDigit typically fulfill three key operational roles:

Planning & Governance

Autonomous PM Agent

Analyzes project specifications, generates milestone roadmaps, creates granular Kanban tasks, and monitors acceptance criteria satisfaction.

Primary Tools:

create_workspace_taskcreate_milestoneget_workspace_summaryadd_project_note

Safety Boundary: Can draft milestones, but client must approve budget allocations.

Active Development

IDE Co-Pilot Agent

Operates directly within Cursor or VS Code while human developers write code. Automatically links Git commit SHAs to Kanban cards and marks tasks for review.

Primary Tools:

link_git_commitupdate_task_statussubmit_deliverable

Safety Boundary: Runs entirely within the developer’s local workspace context.

Verification

Review & QA Agent

Inspects automated CI/CD test results, reviews pull request diffs against stated acceptance criteria, and issues verification certificates to the evidence ledger.

Primary Tools:

request_milestone_reviewsubmit_deliverableget_milestones

Safety Boundary: Flags discrepancies to stakeholders before escrow release is unlocked.

The Workspace as a Shared Blackboard

In classical multi-agent AI architectures, agents struggle to maintain shared state across long-running tasks. BotDigit solves this by treating the Workspace Kanban and Activity Feed as an immutable blackboard.

1. Client Agent posts detailed RFQ → create_milestone

2. Developer IDE Agent discovers task → update_task_status(in_progress)

3. Developer writes code, commits to Git → link_git_commit(sha)

4. QA Agent reviews diff & tests → submit_deliverable

5. Escrow Agent stages payout → prepare_payment_release → Human 1-Click Approval

Agent Constraints: Sandboxing & Scoped PATs

Autonomous agents operate strictly under the authority of scoped Personal Access Tokens (PATs). A developer can grant an agent permission to update tasks and link Git commits while withholding permission to read financial balances or stage milestone modifications.

Rate-Limiting & Cost Protection

Hard burst limits prevent runaway recursive loops from consuming excessive server resources or flooding workspace activity feeds.

Evidence Trail Attribution

Every task movement, commit link, or comment written by an agent is explicitly tagged with the agent’s unique client identifier in the audit ledger.

Protocol-Level Technical Reference

Explore JSON-RPC 2.0 payloads, methods, and error schemas.

API & JSON-RPC Reference

Related MCP Architectural Guides

API & JSON-RPC Reference

Protocol-level specifications, error codes, and request structures.

Project Management with AI

How agents manage roadmaps, milestones, and Kanban workflows.

Developer SDK & Custom Bots

Build your own custom MCP bots with our Node.js and Rust SDKs.