Skip to main content

Five MCP tools

FactLane deliberately keeps its public MCP surface small:

memory_search
memory_get
memory_store
memory_update
memory_status

The purpose of this page is to explain when each tool is useful. The live MCP schema and factlane --help-tools remain authoritative for exact request fields, enums, and signatures.

This is a developer/integrator reference. If you have not connected FactLane yet, start with the Quick start and establish a safe read-only connection first.

Use it to retrieve facts inside one exact scope.

Typical question:

“What verified project facts are relevant to the current release?”

Important boundaries:

  • choose the intended scope;
  • CURRENT and REVIEW_HISTORY serve different purposes;
  • current-state retrieval does not make unverified Candidates eligible merely because they rank highly;
  • bound identity supplied by a trusted host should not be redundantly overridden by the caller.

memory_get​

Use it when you already know which logical memory you want to inspect.

It is useful for understanding provenance, lifecycle and revision details rather than performing a broad relevance search.

memory_store​

Use it to contribute one bounded fact when the launcher profile permits writing.

An ordinary delegated-candidate connection can contribute a Candidate. The tool does not let that agent mint VALIDATED_CURRENT by asserting a privileged verifier identity.

A good contributed fact is:

  • bounded;
  • attributable;
  • scoped;
  • accompanied by freshness policy;
  • idempotent under retry.

It is not an arbitrary transcript or directory dump.

memory_update​

Use it for governed revision operations.

Two important modes are:

  • REVERIFY — trusted verification of the same logical memory identity;
  • REPLACE — a new current revision when the semantic identity changes.

Promotion checks expected record identity and revision so a stale writer cannot silently replace a newer state.

An ordinary delegated agent does not gain verifier authority simply because this tool is visible.

memory_status​

Use it for bounded, read-only observations about a scope's storage and embedding-profile state.

It is not a full monitoring platform, backup facility, or maintenance daemon.

Governed failures​

Expected governed failures use the normal MCP result channel with:

  • status=BLOCKED;
  • a stable error_code;
  • a safe message;
  • empty results;
  • retry metadata where applicable.

Callers should branch on error_code, not scrape exception prose.

Unexpected internal exceptions remain transport errors rather than being relabeled as governed FactLane results.

Example: verify connectivity before writing​

For a global-user status check that does not require project identity:

{"scope":"GLOBAL_USER"}

For a project-scoped current keyword search, a request may look like:

{
  "scope": "PROJECT",
  "project_id": "example-project",
  "intent_class": "CURRENT_PROJECT_STATE",
  "query": "What release constraints are recorded?",
  "retrieval_mode": "CURRENT",
  "retrieval_mode_kind": "KEYWORD"
}

When a trusted host has already bound a corresponding project/workflow/tool identity, follow the live schema and omit conflicting caller-supplied identity.

Next​

  • Architecture — exact request path, identity, storage and recovery boundaries.
  • Environment — runtime and embedding-profile compatibility.