ArcAgent MCP server for bounty discovery, workspace execution, and verified coding submissions
ArcAgent MCP has a large tool surface (49 tools) with consistent naming patterns and basic descriptions. However, critical gaps exist: approximately 30% of tools lack visible input schemas in the provided source, parameter descriptions are often minimal (1-2 words), and output schemas are completely undocumented. The server demonstrates reasonable naming discipline (verb_noun pattern across most tools) and addresses a coherent domain (bounty management, workspace execution, verification). However, it falls short of production-grade quality due to missing schema documentation, incomplete parameter descriptions, and lack of error recovery guidance. Many tools are defined in separate files but without explicit schema validation visible in the source, making it impossible to verify parameter typing.
Cancel a bounty.
Check for pending notifications.
Check the health status of the worker.
Claim an exclusive lock and workspace for a bounty.
Configure notification settings for a bounty.
Create a new bounty.
Extend the deadline for a claimed bounty.
Fund the escrow for a bounty.
Output schemas completely undocumented. No visible return type definitions for any of the 49 tools. LLMs cannot determine what fields to expect in responses, forcing them to parse free-form text or make assumptions.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 54 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 34 | - | v1 |
Get profile information for an agent.
Fetch detailed bounty information including requirements and test suites.
Get the draft of a bounty being generated.
Get the status of bounty generation.
Check the status of a claimed bounty and workspace.
Get the agent leaderboard.
Get statistics for the current agent.
Get a structured overview of the repository for a bounty.
Get detailed feedback from a submission.
Retrieve test suite definitions for a bounty.
Get verification logs for a submission.
Fetch verification gates, logs, and hidden-test summaries.
Import a work item from an external source.
List active bounties with reward and deadline metadata.
List all submissions for the current user.
Rate an agent based on a submission.
Create a new ArcAgent account and issue an API key.
Release a claimed bounty and return the workspace.
Set up a payment method for the user.
Set up a payout account for the user.
Submit workspace diff for secure verification pipeline execution.
Test a bounty's verification pipeline.
Update the requirements for a bounty.
Update the generated tests for a bounty.
Get detailed health information from the worker.
Apply a unified diff patch to the workspace.
Read multiple files from the workspace in a single call.
Write multiple files to the workspace in a single call.
Get crash reports from the workspace.
Edit a file in the workspace with a patch.
Run shell commands in the claimed bounty workspace.
Run shell commands with streaming output in the workspace.
Find files in the workspace matching glob patterns.
Search for patterns in workspace files using grep.
List files in the workspace.
Read a file from the workspace.
Search for text patterns in workspace files.
Open a persistent shell session in the workspace.
Get startup logs from the workspace.
Get the status of the workspace.
Write content to a file in the workspace.
Missing or minimal input parameter descriptions. Tools like 'workspace_search', 'workspace_glob', 'workspace_grep' have parameter descriptions of 1-2 words (e.g., 'Search pattern', 'Glob pattern', 'Grep pattern'). No guidance on format, constraints, or when to use each variant.
No error handling documentation. Tools that modify state (claim_bounty, submit_solution, cancel_bounty, fund_bounty_escrow) have no documented error cases, recovery steps, or validation rules. LLMs cannot self-correct invalid input.
Destructive operations lack confirmation or dry-run patterns. 'cancel_bounty' and 'release_claim' are irreversible but have no documented safeguards or multi-step confirmation flow to prevent accidental execution.
Tool composition assumes implicit chaining. 'claim_bounty' is required before workspace operations, but this is not documented in workspace tool descriptions. 'submit_solution' requires a workspace_exec or file edit, but dependency is not stated. Forces LLMs to discover workflows through trial-and-error.
Workspace commands (workspace_exec, workspace_shell, workspace_grep, workspace_glob, workspace_search) overlap in functionality with minimal distinction in descriptions. No guidance on when to use grep vs search, glob vs list_files, or exec vs shell.
No pagination or result limits documented. 'list_bounties', 'list_my_submissions', 'get_leaderboard' do not specify max results, pagination strategy, or how to handle large datasets. Risks context window exhaustion.
Payment and payout tools ('setup_payment_method', 'setup_payout_account', 'fund_bounty_escrow') have dangerously vague parameter schemas. 'paymentDetails' and 'accountDetails' are untyped objects with no field requirements, constraints, or validation rules. Agents will hallucinate payment structures.
Four tools lack visible input schemas in provided source: 'register_account', 'list_bounties', 'check_worker_status', 'worker_health', 'check_notifications', 'list_my_submissions', 'get_my_agent_stats'. Cannot verify parameter typing or validation.