AI coding agent context for large codebases

Model selection matters less than the harness around the model. ProtoMap is a context engine and model harness for gRPC and Protobuf: it gives Claude Code and Codex organization-wide context, refreshed at development speed, with no full local checkout.

ProtoMap By HoundDog.ai
New chat MessageDispatchCoordination blast radius ···
Ask ProtoMap about your code and data List all gRPC services Which services call <RPC method>? What is the blast radius of changing <field name>?
What breaks if I change MessageDispatchCoordination?
Thinking…
3 graph queries, 1 workspace view
MessageDispatchCoordination has 42 callers and 6 callees, five hops out each way:
Every caller goes through its one method, Dispatch.AuthService is among its 6 callees.Invoicing services, such as CreditNotePublication, call it most.
Shown in the workspace Service MessageDispatchCoordination
What breaks if I change MessageDispatchCoordination? Ask about your code and data…
Opus 5.5 · High
20 services
ServiceMethodsCallers
Service AuthService 4 1,000
Service CartService 3 2
Callers Service ApprovalThresholdAssessment Service CreditNoteAmountCalculation Service CreditNotePublication Service CreditNoteReasonValidation Service InvoiceApprovalCoordination 1–5 of 42 total
42
Repository protos Service MessageDispatchCoordination Method Dispatch
6
Callees Service AccessDecisionCoordination Service AuthService Service CommunicationConsentAssessment Service DeliveryChannelSelection Service MessageDeliveryPreparation 1–5 of 6 total
Service fixture.notifications.v1.MessageDispatchCoordination Methods
MethodRequestResponse
Method Dispatch Message DispatchRequest Message DispatchResponse
Same graph for engineers and agents, via MCP No AI tokens needed
faster per prompt
7.7×
cheaper per prompt
6.4×
fewer tokens
13×

Claude Code with Opus 5.5 at high effort, across 20 prompts on 9 gRPC repositories (5M LOC), with higher accuracy (100% vs. 98%). See the benchmarks

“It works best when Claude has enough starting context to know where to look… Teams that invest in codebase setup see better results.”
Anthropic Engineering • How Claude Code works in large codebases

Model quality is necessary. It is not sufficient.

The same model that ships a clean PR on a 10k-line project struggles on a monorepo. The bottleneck moves from the model to the harness around it: context, permissions, tools, and freshness. That harness decides whether a competent model succeeds on a real codebase or wastes tokens on grep, and it is what an engineering org actually has to build.

What CLAUDE.md and AGENTS.md solve

Conventions, repo orientation, “where the tests live,” coding standards. These are prose guidance an agent reads once. They scale to setup, not to “which 14 services call this RPC.”

Where they fall short

Cross-repo consumer mapping. Field-level usage. Change-impact analysis. The questions an agent has to answer to safely modify a shared API in a system it didn’t write.

The questions behind every cross-service change, answered from code.

Engineers ask in plain English in the web platform. Coding agents ask over MCP. Both get the same deterministic answer from ProtoMap’s context engine.

Multi-hop service dependencies

Which services depend on MessageDispatchCoordination?

Every service that depends on it, directly or through other services, grouped by domain, with the direct callers and their call sites.

Method dependencies

Who calls this RPC, and from where?

Callers and callees of a method two hops out, plus every call site with the calling method and its file location.

Field name dependencies

What is the blast radius of changing credit_card?

The message that carries the field, where it is defined, and the services and repositories that send or receive it.

Unused RPCs

Which RPCs can we clean up?

RPCs with no callers or call sites in the scanned code, plus what to check before you remove each one.

ProtoMap By HoundDog.ai
New chat MessageDispatchCoordination dependents ···
Ask ProtoMap about your code and data List all gRPC services Which services call <RPC method>? What is the blast radius of changing <field name>?
Which services depend on MessageDispatchCoordination?
Thinking…
3 graph queries, 1 workspace view
42 services depend on MessageDispatchCoordination, directly or through others:
Invoicing (18): CreditNotePublication and 17 more.Approvals (9): ApprovalThresholdAssessment and 8 more.Direct callers (5) call Dispatch from 7 call sites.
Shown in the workspace Service MessageDispatchCoordination
Which services depend on MessageDispatchCoordination? Ask about your code and data…
Opus 5.5 · High
20 services
ServiceMethodsCallers
Service AuthService 4 1,000
Service CartService 3 2
Callers Service ApprovalThresholdAssessment Service CreditNoteAmountCalculation Service CreditNotePublication Service CreditNoteReasonValidation Service InvoiceApprovalCoordination 1–5 of 42 total
42
Repository protos Service MessageDispatchCoordination Method Dispatch
6
Callees Service AccessDecisionCoordination Service AuthService Service CommunicationConsentAssessment Service DeliveryChannelSelection Service MessageDeliveryPreparation 1–5 of 6 total
Service fixture.notifications.v1.MessageDispatchCoordination Methods
MethodRequestResponse
Method Dispatch Message DispatchRequest Message DispatchResponse
ProtoMap By HoundDog.ai
New chat CoverShift callers ···
Ask ProtoMap about your code and data List all gRPC services Which services call <RPC method>? What is the blast radius of changing <field name>?
Who calls CoverShift, and from where?
Thinking…
2 graph queries, 1 workspace view
42 methods call CoverShift, all from the facilities services. The first call sites:
FindIncident in building_incident_lookup.go:59SelectResources in emergency_resource_selection.go:59Respond in facility_emergency_coordination.go:73
Shown in the workspace Method CoverShift
Who calls CoverShift, and from where? Ask about your code and data…
Opus 5.5 · High
20 services
ServiceMethodsCallers
Service AuthService 4 1,000
Service CartService 3 2
Callers Method FindIncident Method SelectResources Method AssessEvacuation Method Respond 1–4 of 42 total
42
Service ShiftCoverageCoordination Method CoverShift
2
Callees Method Authenticate Method EnrollLearner 2 total
Method fixture.workforce.v1.ShiftCoverageCoordination/CoverShift Call sites
FromLocation
Method BuildingIncidentLookup/FindIncident building_incident_lookup.go:59
Method EmergencyResourceSelection/SelectResources emergency_resource_selection.go:59
Method FacilityEmergencyCoordination/Respond facility_emergency_coordination.go:73
ProtoMap By HoundDog.ai
New chat credit_card blast radius ···
Ask ProtoMap about your code and data List all gRPC services Which services call <RPC method>? What is the blast radius of changing <field name>?
What is the blast radius of changing credit_card?
Thinking…
2 graph queries, 1 workspace view
PlaceOrderRequest.credit_card goes from 1 sender to 1 receiver, which passes it on:
The frontend sends it to CheckoutService.CheckoutService charges it through PaymentService.Defined in demo.proto:23.
Shown in the workspace Field credit_card
What is the blast radius of changing credit_card? Ask about your code and data…
Opus 5.5 · High
20 services
ServiceMethodsCallers
Service AuthService 4 1,000
Service CartService 3 2
Sent from Service frontend
1
Message PlaceOrderRequest Field credit_card
2
Received by Service CheckoutService Service PaymentService
Field hipstershop.PlaceOrderRequest.credit_card
PropertyValue
Type CreditCardInfo
Definition microservices-demo › demo.proto:23
Blast radius 1 sender · 2 receivers · 1 repository
ProtoMap By HoundDog.ai
New chat Finding unused RPCs to clean up ···
Ask ProtoMap about your code and data List all gRPC services Which services call <RPC method>? What is the blast radius of changing <field name>?
Which RPCs can we clean up?
Thinking…
13 graph queries, 1 workspace view
Two RPCs have no call sites in the scanned code:
RPCCall sitesHandlers
SearchProducts 0 1
Health/Check 0 none found
SearchProducts: check for callers outside the scanned repositories first.Health/Check: platforms, such as Kubernetes probes, call health checks.
Shown in the workspace Method SearchProducts
Which RPCs can we clean up? Ask about your code and data…
Opus 5.5 · High
20 services
ServiceMethodsCallers
Service AuthService 4 1,000
Service CartService 3 2
Service ProductCatalogService Method SearchProducts
0 callers · 0 call sites
Service Health Method Check
0 callers · 0 call sites
Method hipstershop.ProductCatalogService/SearchProducts
CallersCalleesCall sites
0 0 0

Deterministic. No embeddings. No LLM in the analysis loop: the same commit always produces the same catalog. Source code is processed only inside your environment, on the developer’s machine, in your CI, in your Enterprise tenant, or fully on-premises. Only the catalog itself is centrally queryable.

Works with every major AI coding agent

Any agent that implements the MCP protocol.

Once an engineer is ready to apply a change, the coding agent pulls the dependencies directly from ProtoMap over MCP.

Claude Code · run from the repository’s folder
claude mcp add protomap -- protomap mcp serve

Setup guides for every coding agent: docs.protomap.ai

Watch ProtoMap answer the questions agents get stuck on.

HoundDog.ai ProtoMap live demo at MozFest 2026 10:53 Live demo · MozFest 2026

HoundDog.ai ProtoMap Demo at MozFest

Live demo of HoundDog.ai’s ProtoMap at MozFest’s Open Source House: callers and callees, and the blast radius of changing a field name.

HoundDog.ai Benchmarks · Side by side ProtoMap × Claude Code 7× faster · 36× cheaper Claude Code alone6:01 · $14.44 Claude Code + ProtoMap0:50 · $0.40 1:09 Benchmarks · Video

ProtoMap × Claude Code, side by side

The same prompt in two live sessions: ProtoMap’s answers in 50 seconds for $0.40, the baseline’s in 6 minutes for $14.44.

Start free on one machine. Roll out to your organization.

Step 1 · Free

Install it on your machine

Install the self-hosted platform and CLI with Docker, scan your repositories, and query ProtoMap in plain English or over MCP. Free for code scanned on your machine.

  • Self-hosted platform and CLI scanner
  • Bring your own AI provider: Anthropic, OpenAI, Google and others
  • MCP server for your coding agent, including uncommitted changes
Terminal
git clone https://github.com/hounddogai/protomap.git
cd protomap
./self-hosted/scripts/install.sh
Step 2 · Enterprise

Organization-wide context

Enterprise keeps context current across every repository in your organization, so agents no longer clone repositories they don’t own.

  • Cloud hosted or deployed on-premises in your org
  • Direct integrations with GitHub, Bitbucket, and GitLab
  • Runs scans in CI on every pull request
  • Every agent gets context for every repository over MCP

Why not just build another AI agent?

AI agents are great at reasoning. They are an expensive and inefficient way to repeatedly rediscover deterministic facts about your codebase.

Without precomputed context, every agent has to search files, follow references, inspect protobuf definitions, and reconstruct dependencies before it can start on the actual problem. Across developers, prompts, and repositories, you pay for the same discovery again and again in tokens, latency, and compute.

HoundDog.ai separates discovery from reasoning. ProtoMap computes API, field, consumer, and service relationships deterministically, keeps them current as code changes, and makes that evidence available to your existing AI agents through MCP.

  1. 1Static analysis discovers.
  2. 2AI reasons.
  3. 3MCP delivers.

Compute the facts once. Keep them fresh. Let every authorized AI agent reason over them.

Keep AI context fresh with every build

AI coding agents are only as useful as the context they can access. In large codebases, that context changes every time services, APIs, fields, consumers, or dependencies change. HoundDog.ai integrates directly into your CI workflows to continuously regenerate this context from source code.

Code service API method field consumer callsite

Our fast, deterministic analysis engine runs on standard CPU infrastructure with minimal impact on CI latency. ProtoMap computes these relationships once and refreshes them as your code changes.

The resulting evidence can power your service catalog, API and gRPC discovery, microservices documentation, change blast radius analysis, and organization wide context for AI coding agents. Through MCP, authorized agents such as Claude Code can query this continuously updated context whenever they need it.

Use AI where reasoning adds value. Let CI handle the deterministic discovery. The MCP server delivers structured traces, not raw source code, so AI agents reason over facts they no longer have to rediscover.

Frequently asked questions

How does this differ from AGENTS.md and CLAUDE.md?

AGENTS.md and CLAUDE.md hold conventions, repo orientation, and prose guidance an agent should read once. ProtoMap provides structured, queryable context about every API and consumer in your code, refreshed on every commit. The two are complementary: the markdown files tell the agent how to behave, ProtoMap tells it what is in the codebase.

Does this replace retrieval-augmented generation or embeddings?

No. RAG and semantic retrieval are good at finding text that resembles a query. ProtoMap is good at answering deterministic structural questions: which services call this RPC, which fields are consumed where, what breaks if I rename this. The two approaches complement each other.

Which AI coding agents are supported?

Any MCP-capable coding agent: Claude Code, Codex, Antigravity, Copilot CLI, VS Code, and OpenCode.

What happens to my code?

On the free local tier, analysis runs on the developer’s machine and source never leaves it. On Enterprise, scans run inside your tenanted environment (cloud or on-premises), and only the resulting catalog and findings are centrally accessible. Source code never leaves your infrastructure in either tier.

How does the centralized Enterprise model work?

The Enterprise tier connects directly to GitHub, Bitbucket, or GitLab and auto-scans every selected repo. A centralized MCP server keeps an org-wide catalog continuously fresh and serves it to every developer’s agent on demand. No developer needs to clone or check out the full codebase to get useful context, which is the only model that scales to massive monorepos and large microservice estates.

When will GraphQL and REST land?

gRPC is supported today, with GraphQL and REST coming soon.

Try it on your code.

Install the self-hosted version for free. It runs on your machine with Docker, and your source code stays in your environment.

Paste into your coding agent

Set up ProtoMap and show me what it can do. ProtoMap maps gRPC services and protobuf messages across repositories, and gives coding agents that graph over MCP. Docs: https://docs.protomap.ai/quickstart/

Tell me what you are about to do before each step that installs something.

1. Check that this computer has git, Docker with Compose 2.23.1 or newer, at least 4 GB of RAM, and 20 GB of disk space. If anything is missing, tell me and stop.

2. Install ProtoMap Server, in a folder outside this repository:

   git clone https://github.com/hounddogai/protomap.git
   cd protomap
   ./self-hosted/scripts/install.sh

   The installer asks questions. Choose Trial, which runs PostgreSQL locally and logs the CLI in, and keep the default port.

3. Show me the address and the one-time setup key the installer prints. Ask me to open the address, enter the key, and create my organization and owner account. Wait until I say that's done.

4. Scan this repository, from a new shell so the CLI is on the PATH:

   protomap scan .

   If it has no gRPC services in a supported language, tell me, and offer to scan a public example instead:

   git clone https://github.com/GoogleCloudPlatform/microservices-demo.git
   protomap scan microservices-demo

5. From the scanned repository's folder, add ProtoMap to yourself as an MCP server:

   Claude Code: claude mcp add protomap -- protomap mcp serve
   Codex: codex mcp add protomap -- protomap mcp serve
   Antigravity: add "protomap": { "command": "protomap", "args": ["mcp", "serve"] } under mcpServers in ~/.gemini/config/mcp_config.json
   Others: https://docs.protomap.ai/coding-agents/

   Tell me if I need to restart you for it to take effect.

6. Tell me what the scan found, then use ProtoMap's tools to answer one question about my services, such as which services call a method and where. Suggest two more questions I could ask.