Personas

Personas define the "personality" of the interviewer when testing MCP servers. Different personas ask different types of questions, focusing on various aspects like documentation, security, edge cases, or usability.

How Personas Work

A persona shapes how Bellwether interviews your MCP server by:

1. System Prompt - Instructs the LLM how to behave and what to focus on
2. Question Bias - Weights different categories of questions
3. Categories - Defines which types of tests to run
4. Additional Context - Provides specific patterns or examples to try

Built-in Personas

Bellwether comes with four built-in personas:

Technical Writer

id: technical_writer
focus: Documentation and realistic examples
categories: happy_path, edge_case, error_handling

The default persona. Creates comprehensive API documentation with practical examples developers can use as templates.

Best for: Generating documentation, understanding normal behavior

Security Tester

id: security_tester
focus: Vulnerability testing
categories: security, boundary, error_handling

Probes for security vulnerabilities like path traversal, command injection, SQL injection, SSRF, and information disclosure.

Best for: Security audits, finding vulnerabilities

QA Engineer

id: qa_engineer
focus: Edge cases and error conditions
categories: edge_case, error_handling, boundary

Tests boundary values, type coercion, empty values, Unicode handling, and stress conditions.

Best for: Finding edge cases, testing error handling

Novice User

id: novice_user
focus: Usability and error messages
categories: error_handling, happy_path, edge_case

Tests common mistakes new users make and evaluates error message quality.

Best for: Evaluating usability, error message clarity

Question Categories

Each persona weights these categories differently:

Category	Description	Example
happy_path	Normal, expected usage	read_file("/valid/path.txt")
edge_case	Unusual but valid inputs	read_file("file with spaces.txt")
error_handling	Invalid inputs, error conditions	read_file("/nonexistent")
boundary	Limits, extremes, constraints	read_file("/very/deep/.../path")
security	Security-focused tests	read_file("../../etc/passwd")

Using Personas

Via Command Line

# Use a single persona
bellwether interview --persona security_tester npx your-server

# Use multiple personas
bellwether interview --persona technical_writer,security_tester npx your-server

Via Configuration File

# bellwether.yaml
interview:
  personas:
    - technical_writer
    - security_tester

Via Profiles

# Create a security profile
bellwether profile create security --personas security_tester,qa_engineer

# Use the profile
bellwether profile use security
bellwether interview npx your-server

Combining Personas

Run multiple personas for comprehensive coverage:

interview:
  personas:
    - technical_writer  # Documentation
    - security_tester   # Security
    - qa_engineer       # Edge cases

Each persona runs independently, and results are combined in the final report. This provides:

• Complete documentation from technical_writer
• Security findings from security_tester
• Edge cases and error handling from qa_engineer

Persona Output

Each persona contributes different sections to AGENTS.md:

Technical Writer generates:

• Clear tool descriptions
• Parameter documentation
• Usage examples
• Expected return values

Security Tester generates:

• Security considerations
• Vulnerability notes
• Attack surface analysis

QA Engineer generates:

• Edge case documentation
• Error handling patterns
• Boundary conditions
• Limitations

Recommendations

Use Case	Recommended Personas
Quick CI check	technical_writer
Security audit	security_tester
Release testing	All four personas
Documentation	technical_writer, novice_user
Vulnerability scan	security_tester, qa_engineer

Creating Custom Personas

For specialized testing needs, you can create custom personas. See the Custom Personas Guide.

See Also

• Custom Personas Guide - Create your own personas
• interview - Using personas in interviews
• Workflows - Multi-step testing