MCP App Store
by KeenEthics
Developer Tools
Git Diff Patcher Bridge icon

Git Diff Patcher Bridge

by Muhammet Avcı

Overview

Git Diff Patcher Bridge connects ChatGPT to GDP through a privacy-bounded MCP endpoint. It can inspect selected repository context, diagnose patches, read local verification evidence, coordinate durable development state, classify restart requirements, compose private evidence-linked Wire drafts, and create reviewable local requests. The hosted connector does not directly apply patches, execute arbitrary shell commands, force-push code, publish public content, or bypass GDP Desktop authority. Repository-changing actions remain behind GDP Desktop review and local execution boundaries. The connector also supports a seeded Demo Workspace for zero-install reviewer and user evaluation.

Tools

activate_repo_tab

ChatGPT
Ask GDP Desktop to activate one already-open repository by exact repo_id. Path and tab_index inputs are rejected by schema. The immediate result confirms UI-queue admission only; repo_tab_activated activity evidence is the execution receipt.

await_gdp_cycle

ChatGPT
Waits briefly for fresh GDP Desktop automation evidence after a submitted patch or suggested check. This is read-only: it cannot apply patches, run commands, click browser Send or modify files.

await_workspace_chronicle

ChatGPT
Waits briefly for a fresh workspace Chronicle story delta. This is a bounded read-only wait and cannot mutate Agora, execute commands or modify repositories.

build_proof_obligations

ChatGPT
Freeze explicit caller-defined success criteria against one exact operation identity. Read-only: it does not verify, satisfy, execute or mutate any obligation.

build_recovery_plan

ChatGPT
Derive a bounded no-authority recovery recommendation from exact GDP-owned operation evidence and optional pinned semantic comparison.

cancel_operation

ChatGPT
Request exact fail-closed cancellation or admission fencing. A patch-attached suggestion awaiting apply may be cancelled; a standalone queued/delivered suggestion may be blocked for manual review while preserving evidence. Running, terminal, unsupported and ambiguous work is not rewritten.

capture_page_image

ChatGPT
Captures one GDP Desktop allowlisted loopback HTTP page through an isolated exact-origin browser boundary and returns native PNG content without repository writes.

checkpoint_context

ChatGPT
Persists one bounded observable semantic snapshot for an existing durable Lane: constraints, decisions, rejected alternatives, assumptions, open questions, plan, checkpoint and exact next action. It never stores hidden chain-of-thought or grants repository authority. For an exact untouched actorless Browserless bootstrap Task, an explicit checkpoint_id plus important_references=[{type: task_completion, id: complete}] requests retry-safe single-call hosted terminalization: GDP may attest/rebind the current provider chat, persist or reuse that checkpoint and strictly complete that one Lane-owned Task. This compatibility path uses fields already present in older host schemas, so terminal liveness does not depend on a newly exposed callable.

compare_images

ChatGPT
Compares two safe repository PNG/JPEG files without implicit resizing and returns deterministic hash, dimension, pixel, alpha, MAE/RMSE and perceptual metrics.

compile_invocation_plan

ChatGPT
Compiles up to 12 explicit caller-provided calls into one deterministic ordered no-execution plan with validation, previews, blockers and stop conditions. It preserves caller order and never invokes a step, queues work, obtains approval or grants authority.

complete_context

ChatGPT
Terminalizes one exact actorless Browserless Lane-owned Task from the current trusted hosted provider Segment's latest semantic checkpoint. It grants no Browser, command, patch or repository authority. This is a compatibility/convenience path: a fresh trusted hosted provider chat may self-rebind, and the named latest checkpoint may belong to the new current Segment or one of its ancestors. It is not required for liveness when checkpoint_context exposes the task_completion marker path.

compose_newsroom_draft

ChatGPT
Composes one exact candidate returned by inspect_newsroom into a deterministic bounded draft for the internal wire, X, LinkedIn or release notes. Public channels fail closed unless the candidate is public-ready, and release-status wording stays tied to explicit evidence. The hosted gateway only forwards this read-only request to the connected GDP Desktop; it does not save drafts, mutate Agora, call a model or publish externally.

compose_wire_story

ChatGPT
Compose one exact Wire story into a bounded human-readable draft. Public formats fail closed without public-ready evidence and nothing is saved or published.

confirm_upwork_invitation_proposal

ChatGPT
Submit one previously prepared Upwork invitation proposal preview. This is a client-visible external effect. Present the exact preview to the user and obtain explicit approval before calling it. GDP requires the original invitation identity, preview_id, intent_sha256 and confirmed=true; uncertain network outcomes are never auto-retried.

create_notebooklm_audio_overview

ChatGPT
Use the user's already-authenticated Gemini Notebook session in GDP Browser to create or reuse one notebook, add bounded pasted text as a source, and request an Audio Overview. No API key is required. GDP revalidates the exact Browser target between UI transitions and never returns the source text, Google credentials, cookies, arbitrary selectors or browser scripting.

doctor_patch

ChatGPT
Dry-runs a unified diff through GDP Patch Doctor without queueing or applying it.

evidence_diff

ChatGPT
Compare a saved canonical proof baseline with a current GDP-rederived proof. Caller current evidence is never authoritative.

explain_error

ChatGPT
Classifies bounded caller-supplied error text into retry, reconciliation and effect-uncertainty guidance. It does not read private logs, expose hidden reasoning or retry the failed call.

explain_policy_decision

ChatGPT
Explains whether one exact tool request satisfies static visibility, policy and target preconditions. It does not validate every argument, obtain consent, invoke the tool or guarantee execution.

get_selected_gdp_node

ChatGPT
Reads the execution-node and workspace selection stored for the current trusted MCP conversation. During the exact-token shadow cohort only, if the client provides no trusted conversation, MCP session or shadow handle and exactly one eligible node exists, this read may return that node statelessly with no workspace and no persisted selection. Conversation identity is never accepted from model-provided arguments. This hosted read does not mutate the queue, execute work or grant enforced routing authority.

git_diff

ChatGPT
Returns the current diff for the selected GDP repository or selected paths.

git_status

ChatGPT
Returns branch and working tree status for the selected GDP repository.

inspect_agora

ChatGPT
Reads bounded cursor-relative GDP Agora coordination state for the selected repository, including relevant threads, posts, attention and active path or symbol claims. This is read-only.

inspect_agora_commons

ChatGPT
Reads the selected repository’s isolated low-stakes Commons conversations, persistent AI actor profiles and deterministic response invitations. Commons is excluded from task evidence, claims, Chronicle and Newsroom. This read-only tool does not generate speech or mutate state.

inspect_authority

ChatGPT
Separates runtime installation, effective visibility, static policy, open-repository target readiness, user consent and downstream availability for one exact MCP tool. It never invokes the tool or grants authority.

inspect_caller_identity

ChatGPT
Returns a privacy-bounded diagnostic projection of the authenticated hosted MCP caller context. Raw tokens, account IDs, conversation IDs, MCP session IDs, node IDs and workspace IDs are never returned. The result is server-derived and grants no repository, claim, Assignment, Lane, Segment or execution authority.

inspect_continuity

ChatGPT
Reads the selected repository's GDP-owned durable Lane, current Segment, ancestry and latest semantic checkpoint without creating genealogy or granting execution authority.

inspect_git_publication

ChatGPT
Read one exact durable gpr_* Git publication receipt as a privacy-bounded status projection. The result excludes requester/runtime hashes and patch bytes and never stages, commits, pushes, retries, cancels, executes shell commands or mutates repository state.

inspect_image

ChatGPT
Validates one safe repository-relative PNG or JPEG and returns metadata only, without native image content or Base64. Ignored and untracked workspace images are allowed within the same safety limits as read_image. The tool is available only when explicitly enabled by the connected GDP Desktop session.

inspect_limits

ChatGPT
Returns non-secret server limits and, optionally, one effective tool schema and execution-class limits. It does not reserve capacity or bypass quotas.

inspect_newsroom

ChatGPT
Classifies evidence-linked product-news candidates from the durable Agora Chronicle. This deterministic read-only editorial projection reports public readiness, claim scope, warnings and recommended formats. It does not generate final marketing copy, save drafts, post to Agora or publish externally.

inspect_notebooklm

ChatGPT
Inspect already-open Gemini Notebook tabs in the exact live GDP Browser. Returns privacy-bounded runtime identities and readiness only. Google credentials, cookies, arbitrary URLs, selectors and page content are not exposed.

inspect_operation

ChatGPT
Reads one exact typed operation identity from bounded local journal evidence. This forwards a read-only selected-repository request to GDP Desktop and never verifies, retries, cancels or mutates work.

inspect_restart_requirements

ChatGPT
Classify explicit repository-relative changed paths into bounded restart, reconnect, cache-invalidation and Browser Companion reload requirements, together with content-free runtime freshness counters. This is advisory and read-only: it cannot restart GDP Desktop, reconnect the gateway, reload the extension, clear caches or prove lifecycle completion.

inspect_server

ChatGPT
Returns privacy-bounded server identity, effective catalog identity, canonical policy identity, protocol capabilities and non-secret limits. The connected GDP Desktop remains the authoritative executor.

inspect_task_context

ChatGPT
Returns one bounded task snapshot containing repository summary, a deterministic file pack, selected file contents and optional GDP check/queue summaries. Prefer this over several separate discovery calls at the start of a coding task.

inspect_tool

ChatGPT
Returns one exact caller-effective tool descriptor, schemas, annotations, execution policy, authority summary and invocation UI. It never invokes the inspected capability.

inspect_uncertainty

ChatGPT
Derive a bounded uncertainty register from exact GDP-owned operation/proof evidence. No registered uncertainty is not completion proof.

inspect_wire

ChatGPT
Read friendly evidence-linked product stories derived from Agora Newsroom. This is presentation-only and cannot create evidence or publish.

inspect_workspace_chronicle

ChatGPT
Reads a workspace-wide cursor-relative story over durable Agora threads, current task state, attention items and the short-lived GDP pulse. Chronicle is a read model over Agora, not a second source of truth.

list_activity_feed

ChatGPT
Lists recent GDP bridge operational events. This is read-only activity metadata, not hidden model reasoning, and it cannot execute commands or modify files.

list_check_runs

ChatGPT
Lists Auto-Terminal runs with bounded privacy-masked latest-output and latest-failure previews plus compact content-free suggestion and incoming-review state. The same response includes branch, HEAD and working-tree status; do not call repo_summary or git_status unless diff statistics or an explicitly fresh status read is required. Prefer this before separate read_check_run_output, list_suggested_checks or list_incoming_patches calls. After queueing work, use await_gdp_cycle instead of repeatedly polling this or the queue-list tools.

list_files

ChatGPT
Lists files inside the selected GDP repository with optional pattern filtering.

list_gdp_nodes

ChatGPT
Lists up to 100 non-revoked GDP execution nodes owned by the authenticated account, including bounded opaque workspace IDs, labels and active state reported by each Desktop. Local paths, installation identity, credentials, raw capabilities and client-instance identity are never exposed. This hosted control-plane read does not queue repository work.

list_images

ChatGPT
Returns a bounded cursor-paginated content-free repository image inventory with Git state, validation, dimensions, hashes and duplicate groups.

list_incoming_patches

ChatGPT
Lists content-free summaries of queued incoming GDP patches for the selected repository.

list_repos

ChatGPT
Lists repositories currently open as GDP tabs. Use repo_id as the stable session target id; tab_index is only the current visual tab order.

list_suggested_checks

ChatGPT
One-shot detail view for exact Auto-Terminal suggestion items. list_check_runs already includes suggestion status and counts; after submit_patch or submit_suggested_checks, use await_gdp_cycle instead of repeatedly calling this tool. It is read-only.

manage_agora

ChatGPT
Creates or continues one natural GDP Agora team conversation through the authenticated hosted connector. Conversation-only authority; thread status, claims and repository mutation remain unavailable.

manage_agora_commons

ChatGPT
Configures one bounded actor profile or records one budgeted social post or reply. Commons grants no task, release, command, patch or repository authority.

open_chatgpt_tab

ChatGPT
Create one background ChatGPT tab from an exact live GDP Browser source identity. No URL, Send, clipboard or arbitrary navigation authority is exposed.

open_repo_tab

ChatGPT
Asks the connected GDP Desktop to create or reuse one unselected blank repository tab through the native UI queue. It accepts no repository path or filesystem argument. The immediate result proves admission only; physical execution is separately observable through Desktop-owned durable repo-tab receipts.

prepare_upwork_invitation_proposal

ChatGPT
Create Upwork's server-stored preview for one exact client invitation. This does not submit a client-visible proposal. Use it only after the user approved preparing the exact cover letter, bid and attachment choice. The returned preview_id and intent_sha256 are required by the separate confirmation tool.

preview_tool_call

ChatGPT
Produces a deterministic static effect preview after the same bounded validation used by validate_tool_call. It never invokes the target, writes a queue or journal, contacts an external system, obtains approval or claims that predicted effects occurred.

prove_patch

ChatGPT
Classify one exact incoming patch identity from GDP-owned durable full-apply receipts and hash-bound patch verification records. Caller-supplied evidence or authoritative flags are not accepted, and queue presence alone never proves apply or success. This is read-only and cannot apply, run, retry, publish or mutate work.

provision_founding_worker

ChatGPT
Admit one exact read-only Agent Society Founding Worker through the Desktop-owned durable birth controller. Hosted gateway attestation is required; generic local dispatch, generic Send and repository mutation are unavailable.

read_check_run_output

ChatGPT
Reads stdout, stderr and metadata for one Auto-Terminal command the GDP user already ran locally. It never executes commands.

read_context

ChatGPT
Returns a repository or feature-area context pack with a tree, readable files, skipped-file reasons, and suggested follow-up tools.

read_file

ChatGPT
Reads an allowed repository-relative file from the selected GDP repository.

read_files

ChatGPT
Reads multiple allowed repository-relative files from the selected GDP repository in one call, using GDP's configured per-file size limit.

read_folder

ChatGPT
Reads a repository-relative folder as a bounded context bundle, using GDP ignore settings and the configured per-file size limit.

read_image

ChatGPT
Reads one safe repository-relative PNG or JPEG from the selected GDP repository and returns native typed image content. The tool can read ignored or untracked workspace images, but rejects absolute paths, traversal, symlinks, GDP internal state and sensitive locations. It is available only when explicitly enabled by the connected GDP Desktop session.

read_image_region

ChatGPT
Returns one bounded raw-pixel crop from a safe repository PNG/JPEG as native PNG content. The source is never modified and EXIF orientation is not applied implicitly.

read_images

ChatGPT
Reads an explicitly enabled ordered batch of up to 10 safe repository-relative PNG or JPEG files as native typed image content. The batch is all-or-nothing, rejects duplicate normalized paths, preserves order and enforces aggregate byte, pixel and result-size limits.

recall_context

ChatGPT
Reads retained continuity detail on demand without dumping the whole historical conversation. Current kinds expose bounded semantic checkpoints or Segment ancestry.

record_wire_story

ChatGPT
Persist original AI-written presentation copy for one exact current Newsroom candidate. GDP re-resolves the candidate and supplies certainty, claim scope, public readiness and evidence itself; caller prose cannot upgrade those facts. This writes only the private internal Wire presentation ledger and never posts to Agora, changes repository state, approves public copy, schedules content or publishes externally.

recover_gdp_node

ChatGPT
Request fixed-service recovery for one existing GDP node through the configured Windows carrier. No arbitrary command or path. Local ACK is not hosted-online proof.

rehearse_failure

ChatGPT
Re-derive exact GDP recovery evidence and project the fixed failure matrix without creating failures or granting recovery authority.

render_patch_intake

ChatGPT
Renders a ChatGPT Apps UI widget for GDP patch diagnosis or queue status.

render_patch_studio_v3

ChatGPT
Display-only widget. Caller-supplied values do not prove GDP apply, test, commit, push, or live status.

render_society_pulse

ChatGPT
Shows genuinely observed GDP MCP transport events without inventing worker identities or reporting tool traffic as task success. Read-only.

repo_summary

ChatGPT
Returns a compact summary of the selected GDP repository when a local GDP session is connected.

request_git_publication

ChatGPT
Create one durable local publication request for exact selected repository paths. GDP captures trusted local HEAD, branch, upstream, file SHA and staged-index evidence itself. The request is rejected when a selected path is not a current commit-safe change or when an unrelated path is already staged. This tool never stages, commits, pushes, force-pushes or executes shell commands; GDP Desktop retains later execution authority and must revalidate the snapshot before any Git write.

resume_context

ChatGPT
Resumes one durable GDP Lane and returns bounded browserless hydration. With explicit bootstrap_if_missing=true, a trusted hosted provider session may establish the first Task/Lane/Segment. A newly attested provider chat may create one continue Segment under the same Lane. Lineage never carries stale repository, credential, claim, lease or approval authority.

search_repo

ChatGPT
Searches text in the selected GDP repository and returns compact matches.

search_tools

ChatGPT
Searches only the caller-effective MCP descriptor catalog using deterministic name, title, description and input-schema matches. It discovers capabilities without invoking them or expanding authority.

select_gdp_node

ChatGPT
Stores one active account-owned GDP execution node and, when needed, one opaque open workspace as a scoped conversation selection. A sole workspace is selected automatically in global selection mode; multiple workspaces require workspace_id there. During the exact-token shadow cohort only, a client with no trusted conversation, MCP session or shadow handle may observe the sole eligible node statelessly when exactly one exists; that path persists no selection and rejects workspace_id. Scoped shadow clients may keep an omitted workspace node-only for observation, while workspace enforcement still fails closed until a workspace is explicitly selected. Revoked, suspended, foreign-account or unavailable workspace targets fail closed. This control surface does not itself queue, claim or execute repository work.

semantic_idempotency_check

ChatGPT
Compare two exact MCP calls against one pinned caller-effective catalog using GDP validation/preview evidence. This never executes or grants retry authority.

spark_swarm_dispatch

ChatGPT
Dispatch bounded independent prompts to fresh-census Spark workers through the selected GDP Desktop. Dispatch is non-idempotent and must never be blindly replayed after an unknown outcome.

spark_swarm_harvest

ChatGPT
Collect settled outputs from explicitly targeted Spark workers through the selected GDP Desktop. The operation may append a local harvest journal and therefore is not replay-safe.

spark_swarm_status

ChatGPT
Inspect the fresh live Spark worker census and generation state exposed by the selected GDP Desktop. This is observational only and does not dispatch work.

spark_to_chatgpt_bridge

ChatGPT
Harvest selected Spark outputs and inject a bounded synthesis prompt into a GDP Browser ChatGPT tab. This is a UI side effect and must never be automatically replayed.

submit_image_asset

ChatGPT
Validates and stages one image in private GDP state, then creates a durable local review item. Repository files are not modified until explicit GDP Desktop apply.

submit_patch

ChatGPT
Queues one complete unified diff in GDP for local human review. It never applies, commits, rolls back, resets, cleans or pushes changes automatically. Use client_submission_id for retry-safe submissions.

submit_reviewer_reply

ChatGPT
Parses a reviewer reply against a Context Export packet and queues any detected unified diff through GDP Patch Intake for local review. This does not call external services, run commands or apply patches.

submit_suggested_checks

ChatGPT
Queues one bundle of standalone checks or measurements not tied to a newly submitted patch. Put all related standalone commands in the same checks array and one gateway call; do not make one submit_suggested_checks call per command. Put patch-specific checks in submit_patch.verification_plan instead. The bridge stores suggestions only; GDP Desktop keeps terminal execution authority local.

suggest_file_pack

ChatGPT
Standalone planning fallback when inspect_task_context is unavailable or file/check/queue context is intentionally not needed. Do not call it immediately before inspect_task_context because that compound tool already includes the deterministic file pack. This is read-only planning metadata.

transform_image_asset

ChatGPT
Applies a deterministic non-generative image recipe, returns a native preview and creates a durable GDP Desktop Image Asset Review item. Repository files are not modified by the tool call.

validate_tool_call

ChatGPT
Validates one proposed call against the caller-effective descriptor, bounded JSON schema, catalog precondition, static execution policy and exact open-repository target. It does not invoke or queue the proposed tool, obtain consent, reserve capacity or grant authority.

verify_claim

ChatGPT
Verify one narrowly structured caller assertion by re-deriving its proof from GDP-owned operation or patch evidence. Agent identity, prose, confidence and caller-supplied evidence are assertions only and never become authoritative. This is read-only and cannot execute, retry, cancel, apply or mutate work.

verify_operation

ChatGPT
Verify one frozen schema-v2 proof contract against GDP-owned bounded local-journal evidence. Caller-supplied evidence and caller-controlled authoritative flags are not accepted. This is read-only and never executes, retries, cancels, applies or mutates work.

Capabilities

WritesInteractive

Example Prompts

Click any prompt to copy it.

App Stats

91

Tools

3

Prompts

ChatGPT

Platforms

Works with

ChatGPT

Data refreshed daily