Skip to main content
Glama

merestone

Name reserved. This is a placeholder release (0.0.0), published so that the name belongs to the project described below. The first real release — the MCP server — will be 0.1.0. Until then this package does nothing except hold the name; please do not depend on it.

Stable addresses for durable project memory — and the fewest tokens it takes to find them.

merestone is an MCP server that turns a repository's working documents into records addressed by short codes (PL-…, IS-…) that are never renamed and never reused, with every reference re-checked by a gate so a dead pointer fails loudly instead of silently.

The one principle everything else serves

Spend as little context as possible.

Everything this project asks of you — one object, one code; codes that are never reused; records that are superseded rather than edited; references that a gate re-checks — exists for exactly one reason: misunderstanding costs context. Every token spent re-reading a file because a pointer turned out to be dead, or re-deriving what IS-048 meant because the number had been reused, is a token not spent on the actual work.

The rules are not the point. They are the cheapest way we know to keep the context bill down.

mechanism

the context it saves

one record, one code

no tree walk to find "the" document — a code is a few tokens and it resolves

codes never reused, never renamed

no disambiguation between two things that once shared a number

supersede, never edit

no re-reading a whole record just to learn what changed

a gate re-checks every reference

a dead pointer fails at check time, not after three wrong files are loaded

excerpts that say they were truncated

a 40-line body is never mistaken for the whole record

no numbers in tool descriptions

the tool surface stays stable and cacheable

low-frequency commands stay in the CLI

every tool name is permanent context; only the frequent ones earn one

Related MCP server: mcp-project-context-router

Why "merestone"

A merestone is the stone that marks a boundary. It does not move when the land changes hands, and every year the parish walks the boundary and beats the marks — not because the stones rot, but because people forget. This package does the same for a codebase's memory: the code is the stone, the gate is the walk, and the whole thing is written for the next session that arrives without context.

Planned surface

tool

answers

show

what does this code actually say

queue

what is next / what is ready / what is blocked

knowledge

what traps are known before I touch this

touches

which records name this file

claim

which code may I take next

Low-frequency commands (refs, stale, codemap) stay in the CLI on purpose: a tool name is permanent context, so only the frequent ones get one.

Status

  • 0.0.0 — name placeholder (this release)

  • 0.1.0 — first MCP server release (planned)

License

MIT

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      A
      quality
      B
      maintenance
      MCP server that gives coding agents persistent, verified memory of codebase decisions, conventions, and skills, with evidence-based claims that are re-checked via git hooks and human-gated review. Enables memory search, propose/approve, chat harvesting, and critique across MCP-compatible tools.
      21
      87 npm
      1
      MIT
    • A
      license
      C
      quality
      A
      maintenance
      Local-first MCP server that turns project documentation and source code into durable, evidence-backed context for AI agents, with bounded retrieval and explicit gap reporting.
      8
      256 npm
      Apache 2.0
    • A
      license
      Not graded
      quality
      B
      maintenance
      A local-first MCP server that lets coding agents record what they learn as provenance-carrying observations, retrieve and traverse that shared project memory, and keep contradictory claims side by side for human mediation. It enables opt-in peer sync of a workspace across machines, with every assertion keeping its source, basis, and computed trust tier.
      Apache 2.0