SDKs & MCP Forgebench MCP Onboarding agents & tools

Onboarding Agents & Tools

Onboard the agents, MCP servers and tool bindings in your repository in one chat session.

The onboarding steps

Step 1: Start onboarding

In your coding assistant, enter:

Onboard this repo to Forgebench

(Or run the slash prompt /forgebench:onboard_repo)

Your assistant runs a local scanner that looks for agent frameworks (LangGraph, CrewAI, OpenAI Agents SDK, AutoGen, Mastra, Vercel AI SDK) and MCP config files (.mcp.json, .cursor/mcp.json).

In a Java repository it also finds agents built on the Forgebench Java SDK, Spring AI, LangChain4j and Google ADK for Java. A service that builds a Forgebench client or a Spring AI ChatClient is one agent, named after its Maven artifactId or Gradle project. MCP clients declared in Java code are listed as a gap: ask your assistant to add them to the plan.


Step 2: Approve MCP server tool discovery

The scanner finds MCP servers declared in your project but never starts or contacts them without your approval. Your assistant lists them and asks:

Found 2 MCP servers:
1. postgres [stdio: python -m postgres_mcp]
2. github [http: api.github.com]

May I dial postgres to discover its tools?

Reply with the servers to inspect:

Yes, dial postgres. Skip github for now.

The scanner connects to postgres and catalogs its tools (postgres_query, postgres_list_tables, postgres_drop_table).


Step 3: Review the catalog diff & customize

Your assistant summarizes what will be added to your workspace:

Reviewing scan findings:

New Agents:
+ researcher-agent (LangGraph, src/agents/researcher.py)
+ analyst-agent (CrewAI, src/agents/analyst.py)
+ test-fixture-bot (LangGraph, tests/agents/test_bot.py)

MCP Servers & Tools:
+ postgres (3 tools: postgres_query, postgres_list_tables, postgres_drop_table)
  ⚠️ Warning: postgres_drop_table is classified as a DESTRUCTIVE tool.

Proposed Policies:
- Default monthly budget: $50.00 USD per agent
- Default daily key limit: $20.00 USD per agent

What would you like to include, exclude, or customize?

Customize the plan in chat:

Exclude test-fixture-bot.
Set researcher-agent monthly budget to $30.
Proceed with analyst-agent at default.

Things to check before applying:

  • Name collisions: if an agent with the same name exists, Forgebench suggests a unique name such as researcher-agent-2.
  • Test fixtures: exclude any test agents the scanner picked up.
  • Destructive tools: tools that drop, delete or truncate are flagged so you can decide whether to unbind them.

Step 4: Confirm and apply

Your assistant shows the final change plan:

Change Plan [plan_id: plan_883fc2]:
- Register agent 'researcher-agent' [Monthly Budget: $30.00 USD]
- Register agent 'analyst-agent' [Monthly Budget: $50.00 USD]
- Catalog MCP Server 'postgres' [3 tools]
- Bind 'analyst-agent' -> ['postgres_query', 'postgres_list_tables', 'postgres_drop_table']

Shall I apply this plan?

Confirm by saying:

Yes, apply.

Your agents are registered in the workspace catalog in a sealed state.

Change plans expire after 10 minutes. If yours expired, ask your assistant to generate it again.


Next

Budgets & Agent Keys

Claim API keys for your agents and manage spend limits.

Cookie settings

Our Cookie Policy lists every cookie and what it does.