MCP resources, prompts, and playbooks
Frigg publishes machine-readable policy alongside its tools. Clients can inspect these resources to learn the running binary’s support, routing, evidence, and tool-surface contracts without hard-coding them.
Policy and guidance resources
Section titled “Policy and guidance resources”| URI | Media type | Contract |
|---|---|---|
frigg://policy/support-matrix.json |
application/json |
frigg.policy.support_matrix.v4: language and retrieval support matrix. |
frigg://policy/tool-surface.json |
application/json |
frigg.policy.tool_surface.v1: active profile, core/extended manifests, playbook availability, and authoritative source pointers. |
frigg://policy/shell-replacement-map.json |
application/json |
frigg.policy.shell_replacement_map.v1: maps common shell search/read habits to Frigg tools. |
frigg://policy/evidence-packet.json |
application/json |
frigg.policy.evidence_packet.v1: schema and template for claim-to-source evidence. |
frigg://policy/semantic-models.json |
application/json |
frigg.policy.semantic_models.v1: curated semantic defaults and model contracts, not a live leaderboard. |
frigg://guidance/shell-vs-frigg.md |
text/markdown |
Human-readable routing and shell fallback guidance for the active profile. |
frigg://stats/routing |
application/json |
frigg.stats.routing.v1: optional process-local routing counters. |
frigg://policy/evidence-packet.json is a schema/template, not an evidence-composition tool. It explicitly reports not_a_tool=true. Build claims from actual tool outputs and proof reads.
The shell replacement map is guidance, not an additional tool manifest. Always check tools/list before invoking a suggested tool.
Local routing statistics
Section titled “Local routing statistics”Set FRIGG_ROUTING_STATS=1 before starting the MCP server to expose process-local counters through frigg://stats/routing and, where available, workspace.routing_stats.
The snapshot includes whether collection is enabled and counters such as tool calls, zero hits, issued recovery, handle failures, and workspace-gate uses. It is local process state, not cloud telemetry, and resets with the process.
frigg-routing-guide prompt
Section titled “frigg-routing-guide prompt”Frigg exposes one prompt:
| Name | Argument | Contract |
|---|---|---|
frigg-routing-guide |
optional string task |
Returns routing guidance tied to support, semantic-model, tool-surface, shell-replacement, and shell-guidance resources. |
Use it when a host supports MCP prompts and wants current task-oriented routing. The prompt does not execute tools.
Missing resources
Section titled “Missing resources”Reading an unknown URI returns structured data.error_code=resource_not_found. Protocol-version mapping can change the outer JSON-RPC code, so client recovery should branch on data.error_code.
Feature-gated playbook tools
Section titled “Feature-gated playbook tools”Playbook tools require both:
- a binary compiled with
--features playbook; and FRIGG_MCP_TOOL_SURFACE_PROFILE=extendedor its default resolution.
core hides them even when compiled in.
playbook_run
Section titled “playbook_run”Accepts a playbook with playbook_id and ordered steps[]. Each step contains step_id, tool_name, and params. The result is a trace artifact with:
trace_schemaplaybook_idstep_count- ordered step outcomes
An outcome is tagged ok with response_json or err with transport code, message, and optional structured error code.
playbook_replay
Section titled “playbook_replay”Accepts the playbook and an expected_trace_artifact, executes it again, and returns matches, a structured diff, and the replayed artifact.
playbook_compose_citations
Section titled “playbook_compose_citations”Accepts a trace artifact and optional answer text. It returns an answer schema with claims and citations. File citations include tool-call identity, tool name, repository, canonical path, and 1-based source span.
Citation composition uses successful source-bearing trace steps. Error steps and non-file workspace results do not become file citations. Playbook citations remain trace-derived evidence; they do not turn session handles into durable authority.
For normal answers, use evidence and citations with proof reads. Use playbooks only when deterministic trace replay or trace-derived citation payloads are an explicit requirement.
Source Anchors
Section titled “Source Anchors”../frigg/crates/cli/src/mcp/guidance.rs../frigg/crates/cli/src/mcp/routing_stats.rs../frigg/crates/cli/src/mcp/tool_surface.rs../frigg/crates/cli/src/mcp/types/playbook.rs../frigg/crates/cli/src/mcp/server/playbook.rs../frigg/crates/cli/tests/citation_payloads.rs../frigg/crates/cli/tests/rmcp_protocol.rs