Authorized security workflows with signed jobs, evidence and policy-gated AI. Execution fabric for security automation, bug-bounty workflows, and verifiable evidence collection.
This MCP server exposes 15 tools via FastAPI HTTP transport with basic Pydantic schemas. However, critical quality issues severely limit its usability: (1) Tool descriptions are extremely sparse and under 50 characters for many tools, providing insufficient context for LLM selection. (2) Parameter descriptions are generic or missing entirely, most parameters lack actionable guidance on valid formats, ranges, or constraints. (3) Output schemas are completely undocumented, the source shows input schemas but no description of what any tool returns, making chaining and response parsing impossible for agents. (4) No error handling guidance is present; tools lack recovery hints or categorization (retryable vs fatal). (5) Security-critical tools (execute_script, create_execution_grant, create_execution_job) expose policy parameters with minimal constraint documentation. (6) Composition is weak, many tools appear to be low-level building blocks (create_grant → create_job → run_flow) without clear guidance on when to use each or how they chain. The server shows architectural ambition (execution fabric, bounty tracking, AI planning) but lacks the definition rigor needed for reliable agent interaction.
Cancel an ongoing execution run
Request AI-assisted planning within authorization scope
Create a bounty workspace for tracking findings and evidence
Create an execution grant with policy-based authorization
Create a signed execution job from an authorization grant
Execute a script from the catalog with parameters and track execution
Retrieve a signed evidence bundle for a completed run
No output schemas documented for any tool. Agents cannot determine response structure, field names, or types. This makes tool chaining and response parsing impossible.
Parameter descriptions are generic or missing. Complex parameters (scopes, capabilities, limits, snapshot_data, parameters) are typed as array/object with no nested schema, forcing agents to guess structure.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 45 | <=2025-11-25 | v2 |
Get the current status and details of a run
Import a snapshot of findings from a bounty program
List available AI provider descriptors
List available workflows/flows
List available scripts in the catalog
Authenticate user and obtain authentication token
Register a new user account
Execute a defined workflow/flow
Tool descriptions are too short (most under 50 chars) and lack context on: when to use vs similar tools, side effects, prerequisites, error recovery.
No error handling guidance. Tools like execute_script, create_execution_job, and run_flow have no descriptions of failure modes, retryability, or recovery steps.
Composition weakness: Tools like create_execution_grant → create_execution_job → run_flow suggest a complex multi-step flow, but no guidance explains the relationship, why each step is separate, or typical invocation patterns.
Security-critical parameters lack validation constraints. scopes array in create_execution_grant mentions 'host, url_prefix, cidr, path, opaque' in description but no enum or nested schema. capabilities array similarly vague.
Platform enum in create_bounty_workspace is missing. Description does not document valid platform values (hackerone, bugcrowd, intigriti, yeswehack?, custom?), forcing agent to guess.
Insufficient guidance on resource discovery. list_scripts and list_flows have no pagination parameters (limit, offset, cursor) or explanation of how to filter/search, limiting agent usability.
No explanation of state machine or workflow. What is the relationship between grant, job, and run? Can a job exist without a grant? Can a flow be run without a job? These questions are unanswered.