MCP Referenceinvarium_upload_blueprint

invarium_upload_blueprint

invarium_upload_blueprint

Upload an agent blueprint to the Invarium dashboard. Takes a JSON string describing your agent's tools, workflows, and constraints. Use the agent blueprint template to generate this JSON from your codebase.

Parameters

NameTypeRequiredDescription
blueprintstringrequiredJSON string of the agent blueprint.
agent_namestring | nulldefault: nullAgent name. If not provided, extracted from the blueprint.

Returns

Confirmation with tool count, workflow chains, confidence level, and dashboard link.

Example

Blueprint uploaded successfully
  Agent: customer-support-agent
  Tools: 3 detected
  Workflow chains: 2 identified
  Confidence: high
  Dashboard: https://app.invarium.dev/agents/customer-support-agent

Usage

Pass your agent blueprint as a JSON string. The blueprint must include at minimum an agent_name (or pass it as the second parameter) and at least one tool.

invarium_upload_blueprint(
  blueprint='{"agent_name": "my-agent", "tools": [{"name": "search", ...}]}',
  agent_name='my-agent'
)

If a blueprint already exists for the given agent name, it will be replaced with the new version. Previous test runs are preserved.

See the Blueprint Schema for the full list of supported fields, and Blueprint Examples for practical templates.


Validation rules

The blueprint is validated on upload. The following rules apply:

  • agent_name is required (either in the blueprint JSON or as the agent_name parameter)
  • At least one tool must be defined in the tools array
  • Each tool must have a name and description
  • framework must be one of: langchain, crewai, autogen, openai_agents, openai, custom
  • side_effects on tools must be one of: none, read, write, delete
  • error_handling on tools must be one of: retry, fallback, fail, ignore

If validation fails, the error message will indicate which field is invalid and what values are accepted.


Confidence scoring

After uploading, Invarium analyzes the blueprint and assigns a confidence level that indicates how well the blueprint describes the agent:

ConfidenceMeaning
HighBlueprint includes tools with parameters, return types, constraints, and workflow chains. Invarium can generate highly targeted tests.
MediumBlueprint includes tools and some constraints, but is missing details like parameters, side effects, or workflow chains. Tests will be generated but may be less targeted.
LowBlueprint is minimal — only agent name and basic tool names. Invarium will generate generic behavioral tests. Adding more detail will improve test quality.

Error responses

ErrorCauseFix
Validation error: agent_name is requiredNo agent name in blueprint or parameter.Add agent_name to the blueprint JSON or pass it as a parameter.
Validation error: at least one tool is requiredThe tools array is empty or missing.Add at least one tool definition.
Validation error: invalid framework valueThe framework field contains an unsupported value.Use one of: langchain, crewai, autogen, openai_agents, openai, custom.
Blueprint too largeThe blueprint JSON exceeds the maximum size (500 KB).Reduce the blueprint size — consider splitting into multiple agents if needed.
Was this page helpful?