get_repo_memory

Instructions

Implementation Reference

• src/repo_memory/mcp_server.py:19-28 (handler)

  The tool handler function for 'get_repo_memory'. It is decorated with @mcp.tool() (registration) and calls store.render() to produce a Markdown document of the repo's .ai-memory/ contents.

  @mcp.tool()
  def get_repo_memory(fact_limit: int = 50) -> str:
      """Return the entire `.ai-memory/` of the current repo as a Markdown document
      ready to drop into your LLM context. Call this **before** starting any task
      in this repo so you don't redo work other agents already verified.
  
      Args:
          fact_limit: cap on number of facts (default 50, most recent first).
      """
      return store.render(_REPO_ROOT, fact_limit=fact_limit)

• src/repo_memory/store.py:257-298 (helper)

  The store.render() function called by the handler. It gathers facts, decisions, and gotchas from .ai-memory/ and assembles them into a single Markdown string.

  def render(root: Path, *, fact_limit: int | None = 50) -> str:
      """Render the whole .ai-memory/ as one markdown document for an LLM prompt."""
      if not is_initialized(root):
          return "_(no .ai-memory yet — run `repo-memory init`)_"
  
      parts: list[str] = []
      parts.append(f"# Memory snapshot for `{root.name}`\n")
      parts.append(f"_Rendered at {_now()}._\n")
  
      facts = list_facts(root, limit=fact_limit)
      parts.append(f"## Facts ({len(facts)})\n")
      if facts:
          for f in facts:
              ev = f.evidence or {}
              ev_str = ""
              if ev.get("file"):
                  lines = ev.get("lines")
                  ev_str = f" — `{ev['file']}`{':' + str(lines) if lines else ''}"
                  if ev.get("verified_at"):
                      ev_str += f" (verified {ev['verified_at']})"
              tags = " ".join(f"#{t}" for t in f.tags)
              parts.append(f"- **{f.claim}**{ev_str}  {tags}".rstrip() + "  _\\<{}>_".format(f.id))
      else:
          parts.append("_(empty)_")
  
      parts.append("\n## Decisions\n")
      decs = list_decisions(root)
      if decs:
          for p in decs:
              parts.append(f"### {p.stem}\n")
              parts.append(p.read_text(encoding="utf-8").strip() + "\n")
      else:
          parts.append("_(none yet)_")
  
      parts.append("\n## Gotchas\n")
      gotchas_path = _memdir(root) / GOTCHAS_FILE
      if gotchas_path.exists():
          parts.append(gotchas_path.read_text(encoding="utf-8").strip())
      else:
          parts.append("_(none yet)_")
  
      return "\n".join(parts) + "\n"

• src/repo_memory/mcp_server.py:19-19 (registration)

  The @mcp.tool() decorator registers 'get_repo_memory' as an MCP tool with the FastMCP server.

  @mcp.tool()

• src/repo_memory/mcp_server.py:20-26 (schema)

  The function signature and docstring define the schema: accepts an optional 'fact_limit' (int, default 50) and returns a string.

  def get_repo_memory(fact_limit: int = 50) -> str:
      """Return the entire `.ai-memory/` of the current repo as a Markdown document
      ready to drop into your LLM context. Call this **before** starting any task
      in this repo so you don't redo work other agents already verified.
  
      Args:
          fact_limit: cap on number of facts (default 50, most recent first).