Fuse
Reference

Resources Reference

The fuse:// URI templates an AI client can read for semantic context using default options.

The Fuse MCP server exposes its read workflows as resources in addition to tools. MCP (Model Context Protocol) is the open protocol that lets an AI client call external tools; a resource is the passive counterpart, a read addressed by a URI rather than a call with arguments. A client reads a fuse:// URI and receives the result produced with default options against the persistent semantic index. This page documents the URI templates and what each returns, plus the playbook prompts.

This page is for engineers wiring resource reads into a client and for anyone deciding between a resource and the equivalent tool. It also documents the playbook prompts, the selectable plans that teach the verified-edit loop.

Purpose and scope

Resources read the persistent index (building it on first use) and return the result as the resource content. They use default options: no token limit, default render tiers, and a depth the workflow chooses. When a task needs a token budget, a render format, seeds of several kinds, or session elision, prefer the equivalent tool in Tools reference, which accepts those parameters. Use a resource when the default behavior is sufficient and a URI is more convenient than a tool call.

In the templates below, {path} is the workspace directory and the remaining segments supply the query.

URI templates

URI templateReturnsMirrors
fuse://map/{path}A workspace map: indexed symbols, routes, and countsfuse_workspace action=map
fuse://status/{path}The workspace status: index mode, verification grade, freshnessfuse_workspace action=status
fuse://localize/{path}/{query}Ranked candidate files and symbols for the task, no source bodiesfuse_find kind=task
fuse://context/{path}/{seed}Planned and emitted context (source bodies, manifest, provenance) for a named seedfuse_context
fuse://review/{path}/{since}The semantic impact of a change since the git ref: changed files, blast radius, packed contextfuse_review
fuse://ledger/{path}/{session}A session's accumulated graded claims (the evidence trail)fuse_impact (with a session)
fuse://diff/{path}/{session}The diagnostics a session's edits introduced or resolved since its baseline (read-only)fuse_check --delta
fuse://diagnostics/{path}/{session}A session's whole-state diagnostics (live resident set, or the recorded baseline)fuse_check --full

The {seed} segment of the context resource is treated as a symbol seed; for service, request, route, or config seeds, or to pass several seeds, use the fuse_context tool. All resources return content with the MIME type text/plain. The session diff and diagnostics resources are read-only: unlike fuse_check, they never establish a baseline, so reading them is idempotent; they report a note when no resident workspace serves the root or the session has no baseline yet. See claim grades for the ledger and its grades.

Playbook prompts

The server also exposes five playbook prompts: selectable, anchored plans that expand into an ordered walk of the eight-tool loop, so a model gets the phase order without a separate skill document. Select one by name and supply its anchor.

PromptAnchorPlan
fix-build-errora diagnostic idlocate the site, check the fix at compiler grade, impact any changed signature, review
implement-featurean issue, optionally a git baselocalize, read context, edit and check each file, impact, test, review
review-pra merge basereview the diff, impact the changed signatures, run covering tests, handoff
rename-symbola symbol's fully qualified nameimpact, verify-gated rename via fuse_refactor, apply, review
add-endpointa route prefixresolve the route neighborhood, read context, add and check the handler, resolve wiring, test, review

What this does not cover

This page documents the resource URIs and their default behavior. It does not cover the parameters that set token budgets, render format, seed kinds, or session ids, which are available only on the tools. See Tools reference for those.

Next

Continue to Tools reference when you need a token budget, a render format, or multi-seed context on a read.

On this page