donaldfilimon/abi / .agents/skills/abi-mcp-transport
abi-mcp-transport
VerifiedValidate ABI MCP JSON-RPC stdio and its custom loopback HTTP compatibility endpoint, including auth, limits, notifications, and shutdown.
ABI MCP transport
Use this skill for the real abi-mcp transport surface. ABI provides
JSON-RPC 2.0 over stdio and attempts a loopback-only custom HTTP compatibility
listener at startup. A bind failure leaves stdio running. The HTTP path serves
persistent MCP 2024-11-05 HTTP+SSE sessions (not Streamable HTTP, 2025-03-26);
the process itself is not a long-lived daemon.
Real launch paths
Build and start stdio:
./tools/cargo.sh build -p abi-mcp
./mcp/launcher.sh stdio
The server exits on stdin EOF. List or call the frozen tools by sending normal
JSON-RPC requests over stdio; there are no /abi-mcp-transport commands.
Configure the automatically attempted custom loopback listener:
ABI_MCP_HTTP_PORT=8080 ABI_MCP_HTTP_TOKEN=local-secret \
./target/debug/abi-mcp
GET /sseopens a persistent session:event: endpointnames/message?sessionId=<id>(a random v4 UUID) and the stream stays open on its own thread (crates/abi-mcp/src/sse.rs).POST /message?sessionId=<id>answers202with an empty body and publishes the JSON-RPC response asevent: message, data byte-identical to stdio. Notifications queue no event. An unknown or closed session gets404before dispatch.- At most 16 sessions (
503beyond); a session ends on client disconnect or server stop, with a:keepalive after 15 s idle. POST /messagewithout asessionIdkeeps the one-shot compatibility mode: one JSON-RPC message per connection, response directly over HTTP (200), and accepted notifications get202with an empty body.- A present Origin must identify
localhostor127.0.0.1over HTTP(S). - The listener binds only to loopback.
ABI_MCP_HTTP_PORT=0, empty, malformed, or out-of-range environment values
fall back to 8080. Direct Rust tests may still use HttpConfig { port: 0 } for
an ephemeral listener.
Enforced protocol boundaries
| Boundary | Behavior |
|---|---|
| Stdio physical line | 64 KiB maximum; one parse error, discard through newline, then recover |
| JSON nesting | 32-container maximum before JSON parse |
| String/field input | Middleware applies its bounded field checks |
| Notification | Omitted id dispatches without a JSON-RPC response |
Explicit id: null | Invalid request; omitted and null remain distinct |
| HTTP auth | Optional exact Bearer token via ABI_MCP_HTTP_TOKEN |
| HTTP+SSE session | Unknown/closed session 404 before dispatch; more than 16 open 503 |
| Shutdown | EOF wakes and joins the actual bound HTTP listener |
Frozen tool catalog
The twelve contract-tested tools are ai_run, ai_complete, ai_learn,
ai_train, wdbx_query, scheduler_stats, scheduler_info,
connector_test, gpu_status, plugin_list, wdbx_stats, and plugin_run.
Do not change the catalog or tests/golden/mcp-tools-list.json as incidental
transport cleanup.
Validation
ABI_WDBX_PATH=:memory: ABI_WDBX_PERSIST=0 ./tools/cargo.sh test -p abi-mcp
./tools/cargo.sh clippy -p abi-mcp --all-targets -- -D warnings
./tools/cargo.sh build -p abi-mcp
Tests and smokes must never open the user's live ~/.abi store.
Claim boundary
- JSON-RPC stdio and the frozen twelve-tool catalog are contract-tested.
- The loopback HTTP surface serves persistent MCP 2024-11-05 HTTP+SSE sessions,
contract-tested against stdio bytes (
transport_contract.rs), plus the one-shot direct-replyPOST /messagemode. - It is not Streamable HTTP (2025-03-26), WebSocket/gRPC streaming, a production non-loopback service, or a claim of TLS/authz/rate-limit completeness.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |