Use cases

Read from every tool you code in. Answer in every assistant you use.

Golden Gate

reads your tools

Golden Gate Bridge

serves anything that speaks MCP

© OpenStreetMap contributors
  1. 01 via Chat assistant

    A weekly review in one question

    Standups and weekly updates start with remembering. Your work arrives grouped by project, across every tool.

    Readsrecent_activitylist_projects

    Share the projects you would report on. Leave client work out.

    What did I work on this week?

    Readsrecent_activitylist_projects
    • pharmfindr: rate-limited review submissions, added gRPC health checks, cursor pagination (PR #42 open)
    • forge: token usage on every message
    • harbinger: MLFQ cuts p99 latency from 41.7 to 18.2 ms at 5k msgs/s

    Sample data

  2. 02 via Coding agent

    Start in one tool, finish in another

    Your coding agent can read what a different tool did in the same repository, so a new session starts with context.

    Readsget_project_context

    Same folder, two tools, one project. Worktrees are filed under their repository.

    ~/code/pharmfindr
    $ codex "finish the review rate limiter"
reads get_project_context("~/code/pharmfindr")
Found it. Claude Code started this 2h ago:
  api/reviews.go  rate limit added, tests pass (3d8f6c1)
Still open: PR #42, no merge seen.
Picking up in api/reviews.go.
  3. 03 via Scheduled agent

    A Monday nudge for stale pull requests

    A scheduled agent checks what you opened and never saw merged, and posts it where you will see it.

    Readslist_pull_requests

    Status comes from what the session showed. It does not ask GitHub.

    Weekly agent

    Open pull requests worth a look:

    • pharmfindr #42 opened 2 days ago, no merge seen
    • harbinger #17 was only mentioned. Nothing was opened.
    list_pull_requests
  4. 04 via Email draft

    The Friday update, drafted

    A changelog or status email starts from the commits and issues that actually happened, with the evidence behind each.

    Readslist_commitslist_issues

    Issue titles and states are sent. Descriptions and comments never are.

    Draft
    To
    team@example.com
    Subject
    Shipped this week: pharmfindr

    Hi all, what went out:

    • Fixed the staging OAuth redirect (a41c9e2)
    • Cursor pagination for search (7be0d13)
    • Rate limit on review submissions (3d8f6c1)
    • Filed #51: rate-limit review spam
  5. 05 via Voice assistant

    Ask out loud on the commute

    Any voice assistant that can call MCP tools can answer from your recent work without you opening a laptop.

    Readsrecent_activity

    Answers are only as private as the assistant you ask. The token can only read.

    “What did I ship this week?”

    Three things. Pagination for the CDMO search, a rate limit on reviews, and the staging redirect fix.

  6. 06 via Document

    A performance review without the scramble

    Pull a year of pull requests and commits into highlights, while the work is still on your mind and in the record.

    Readslist_pull_requestslist_commits

    Looks back up to 365 days. Only what you chose to share.

    Highlights, this year

    Built from merged and created work:

    • Shipped cursor pagination for CDMO search (PR #42)
    • Cut MLFQ p99 latency from 41.7 to 18.2 ms at 5k msgs/s
    • Added per-message token usage to Forge
  7. 07 via Calendar assistant

    Walk into standup ready

    A calendar assistant can attach yesterday, today and what is still open to the invite, ten minutes before.

    Readsrecent_activityget_project_context

    Reads summaries, never transcripts.

    10:00Team standup

    Yesterday

    • Rate limit on review submissions
    • gRPC health checks

    Still open

    • PR #42 awaiting merge
    • Issue #51 filed
  8. 08 via Editor agent

    Don't solve it twice

    Before an agent starts, it can search your past work and notice you already fixed something like this.

    Readssearch_workget_session

    Search covers titles and summaries. Raw transcripts never leave your machine.

    auth/redirect.go
    12func redirectURI(env string) string {
13    // staging callback is stale
14    return baseURL + "/callback"
15}

    You fixed a staging redirect like this 30h ago (a41c9e2, auth/redirect.go). Reuse that approach?

  9. 09 via Chat assistant

    Bring your chat history along

    Export claude.ai or ChatGPT and those conversations become projects beside your code, each shared by its own rule.

    Readssearch_workget_session

    Artifacts and uploads travel as metadata only: id, title and status.

    What did I decide about the database schema in my earlier chats?

    Readssearch_workget_session
    • AI claude.ai, “Reviews schema”: ratings get their own table, one row per review.

    Sample data

  10. 10 via Chat assistant

    Keep client work out of it

    Share the projects you choose and deny the rest. A deny rule always wins, even over a broader allow.

    Readslist_projects

    Denied sessions are dropped before they are parsed, and their names are never logged.

    What did I do for client-x this week?

    Readslist_projects
    • I do not see a project called client-x. Only projects you chose to share are available to me.

    Sample data

Who it fits

  • You work across several AI coding tools and lose track of what you did where.
  • You want an assistant, agent or app to answer questions about your own recent work.
  • You are comfortable running one small server, and want to decide exactly what is on it.

What it is not

  • A hosted service. There is none: one deployment serves one person, or a small team that trusts each other.
  • Something that acts. Golden Gate Bridge only reads. Scheduling, posting and sending is done by the tool you connect.
  • Live status. Sessions are summarized when you sync.

Every tool is read-only, and the token cannot write or delete. See the MCP tools, Security, or the repository.