feat(control-plane): embedded MCP server at /mcp (zero-setup harness integration)

[AbirAbbas]

What & why

AgentField had no MCP server — the only way for an AI harness to drive installed
agents was the af CLI plus the agentfield-use skill. This PR ships an
MCP server embedded in the control plane itself, served on the same port as
the REST API. Whether the control plane was started by the CLI (af server)
or the desktop app, MCP is just there — zero extra processes, zero setup. A
harness connects with one command:

claude mcp add --transport http agentfield http://localhost:8080/mcp

It is on by default, with an env kill-switch: AGENTFIELD_MCP_ENABLED=false
makes /mcp return 404. The endpoint sits behind the same global API-key auth
and trust domain as /api/v1, and the tools call the existing discovery /
execution service layer directly (no loopback HTTP).

• Transport: streamable HTTP, JSON-RPC 2.0 over POST /mcp (stateless;
  Mcp-Session-Id / MCP-Protocol-Version headers tolerated and ignored).
  GET /mcp → 405, OPTIONS /mcp → 204.

The five tools

Tool	Purpose
discover_agents	List agents + reasoners (id, description, entrypoint tags). Defaults to active-only; health:"all" includes unhealthy.
get_reasoner_schema	Input/output JSON Schema for a reasoner, as served by discovery.
execute_reasoner	Validate node.reasoner exists, start an async run, return {run_id, status:"accepted"}.
get_run	Run status + result/error + per-execution summaries.
wait_run	Server-side poll to terminal or timeout (default 60s, hard-capped at 120s); adds a timed_out flag.

Validation Contract → tests

• initialize returns protocolVersion + capabilities.tools + serverInfo →
  TestMCP_InitializeAndToolsList (also asserts unknown protocol falls back to
  2025-06-18 and serverInfo carries the build version).
• tools/list returns exactly the 5 tools, each with a valid JSON Schema →
  TestMCP_InitializeAndToolsList.
• discover_agents default returns only active agents; health:"all" includes
  dead ones with health_status → TestMCP_DiscoverAgents.
• execute_reasoner bad/malformed target → tool error naming discover_agents;
  good target → run_id and a created execution → TestMCP_ExecuteReasoner.
• get_run aggregates status/result/error + per-execution summaries →
  TestMCP_GetRun.
• wait_run respects the timeout cap and returns timed_out on a never-finishing
  run (and returns immediately for a terminal run) → TestMCP_WaitRun.
• Unknown method → -32601; unknown tool → -32602; batch arrays rejected →
  TestMCP_ProtocolMiscMethods.
• AGENTFIELD_MCP_ENABLED=false → /mcp is 404; default → enabled;
  GET /mcp → 405 → TestMCPRoutes_DisabledReturns404,
  TestMCPRoutes_EnabledByDefault.
• Config default-on + env override → TestMCPConfig_IsEnabledDefault,
  TestApplyEnvOverrides_MCPEnabled.

Test results

• go build ./... — clean
• go vet ./internal/config/... ./internal/handlers/... ./internal/server/... — clean
• go test for the config, handlers, server (MCP) and skillkit packages — all pass
• gofmt -l on changed files — empty
• golangci-lint — no issues in any changed/new file

One pre-existing, environment-specific test (TestStartAndStopCoverAdditionalBranches,
admin gRPC listener bind) fails identically on clean origin/main in this sandbox
and is unrelated to this change.

🤖 Generated with Claude Code

[github-actions]

📊 Coverage gate

Thresholds from .coverage-gate.toml: per-surface ≥ 84%, aggregate ≥ 85%, max per-surface regression ≤ 1.0 pp, max aggregate regression ≤ 0.50 pp.

Surface	Current	Baseline	Δ
control-plane	86.80%	87.40%	↓ -0.60 pp	🟡
sdk-go	92.40%	92.00%	↑ +0.40 pp	🟢
sdk-python	93.82%	93.73%	↑ +0.09 pp	🟢
sdk-typescript	91.05%	90.42%	↑ +0.63 pp	🟢
web-ui	84.76%	84.79%	↓ -0.03 pp	🟡
aggregate	85.51%	85.75%	↓ -0.24 pp	🟡

✅ Gate passed

No surface regressed past the allowed threshold and the aggregate stayed above the floor.

[github-actions]

📐 Patch coverage gate

Threshold: 80% on lines this PR touches vs origin/main (from .coverage-gate.toml:thresholds.min_patch).

Surface	Touched lines	Patch coverage	Status
control-plane	578	84.00%	✅
sdk-go	0	—	➖ no changes
sdk-python	0	—	➖ no changes
sdk-typescript	0	—	➖ no changes
web-ui	0	—	➖ no changes

✅ Patch gate passed

Every surface whose lines were touched by this PR has patch coverage at or above the threshold.