MCP Server

Dapper includes an MCP server that exposes debugger operations as tools for AI agents.

The server connects to a Dapper proxy session and translates tool calls into debug operations such as listing threads, reading stack frames, inspecting variables, setting breakpoints, and navigating execution.

Start The Server

dapper mcp

When exactly one session is active, Dapper can target it automatically. For multiple sessions, pass the control port:

dapper mcp --control-port=47823

Choose A Toolset

Use the default standard toolset for normal agentic debugging:

dapper mcp --toolset=standard

Use full when the agent needs evaluation, variable mutation, or memory access:

dapper mcp --toolset=full

Use raw only when the agent needs to send adapter-specific DAP requests that are not represented by typed tools.

JSON Output

Pass --json when the consumer parses tool results rather than reading them:

dapper mcp --json

Results use the same envelope as the CLI, with the response under result and any session context under context. Errors render as {"error": "..."} so a client can parse success and failure the same way.

A few tools are exempt: memory reads and writes stay textual, and raw DAP requests, thread snapshots and config return their own JSON payload with no envelope. See dapper mcp for the full list.

The flag is global, so dapper --json mcp works too, and DAPPER_OUTPUT_JSON=true or output_format = "json" in Dapper's config.toml sets the same thing. The env var is a strict boolean, so a value like 1 aborts the command rather than enabling JSON.

Session Targeting

For long-lived agent sessions, pass a stable scope through DAPPER_SCOPE_ID when your agent environment provides one:

DAPPER_SCOPE_ID=my-agent-session dapper mcp

The scope helps the proxy and agent pair automatically, but --control-port is still the most deterministic option when several sessions are active.

For the exact toolset contents and custom tool selection flags, see dapper mcp.