How much does Morph cost?

Per token pricing on a credit balance. The free tier covers 200 requests a month, pay as you go starts with a $10 top up with $5 added to the first one, and Scale is $200 a month for 40 million credits worth $400. Enterprise is custom, with SSO, custom rate limits and priority support. morph-v3-fast costs $0.80 in and $1.20 out per million tokens, morph-v3-large $0.90 and $1.90, Compact $0.20 and $0.50, and the Router $0.005 per request. Credit purchases carry a 5.5 percent fee with a $0.80 minimum, and startups can get up to $1,000 in credits.

Does Morph support MCP and tool calling?

Full under the Agentic Index verification standard. Morph publishes public evidence of Model Context Protocol (MCP) support, tool calling, and API or SDK extensibility for connecting agents to external tools. Evidence basis: The API at api.morphllm.com speaks the OpenAI chat completions format and the Anthropic messages format, with further routes for compaction, routing, embeddings and rerank, and authenticates with a bearer token or an API key header. A TypeS. Graded from public evidence only against the Agentic Index 14 point capability taxonomy. No vendor pays for placement.

Does Morph have SOC 2, SSO and role based access control?

Full under the Agentic Index verification standard. Morph publishes public evidence of enterprise security and identity controls, covering a compliance posture such as SOC 2 or ISO 27001, single sign on (SSO or SAML), and role based access control (RBAC). Evidence basis: Organizations have admin and member roles, organization and personal API keys, and monthly budgets per member that admins set and that pause the member's keys when reached, with SSO on Enterprise. Morph holds a SOC 2 Type I report, shown on. Graded from public evidence only against the Agentic Index 14 point capability taxonomy. No vendor pays for placement.

Can Morph be self hosted or run in a private VPC?

Full under the Agentic Index verification standard. Morph publishes public evidence of self hosted or private VPC deployment, along with data residency options for controlling where agent data is processed and stored. Evidence basis: Deployment runs in Morph's cloud or the customer's: Fast Apply can be self hosted, including air gapped, and on premises and private cloud deployments are sold through the sales team. Dedicated inference runs on B200 or B300 GPUs. Standard. Graded from public evidence only against the Agentic Index 14 point capability taxonomy. No vendor pays for placement.

Back to vendors

Morph

Q: Who is Morph for?

companies building their own coding agents, app builder and vibe coding platforms, developers running Claude Code, Codex, Cursor or OpenCode on large codebases, AI platform teams cutting model cost inside agent loops

Also known as: MorphLLM, morphllm.com, Morph Fast Apply, Morph WarpGrep, WarpGrep, AutoInfra, Inc.

Visit site

Agent infrastructureindependentVerified 2026-11-08

Entry pricePay as you go from a $10 top up, billed per token, with 200 free requests a month. Scale is $200 a month for $400 of credits. Enterprise is custom.Full pricing detail

Morph builds fast, specialized models that sit inside other companies' coding agents. Fast Apply merges an agent's partial code edits into full files at more than 10,500 tokens a second, WarpGrep is a code search subagent that finds the right files and lines in its own context window, and Compact, Reflex and a model Router trim context, make quick decisions and pick models. It also hosts open weight models such as Kimi and DeepSeek behind one API that speaks both the OpenAI and Anthropic formats.

Morph, sold by AutoInfra, Inc. of San Francisco and a Y Combinator Summer 2023 company, builds inference for coding agents. It sells no agent of its own to end users. Its customers are the teams building coding agents and the developers running them, and its pitch is that the slow, repetitive steps inside an agent's loop, applying edits, searching a codebase and trimming context, run faster and cheaper on small specialized models than on a frontier model.

Fast Apply, on the morph-v3-fast and morph-v3-large models, takes the partial edits an agent writes, with placeholder comments for unchanged code, and merges them into full files at more than 10,500 tokens a second. WarpGrep is a code search subagent: it runs grep, file reads and directory listings in its own context window and hands back the file and line ranges the main agent needs, over a local repository or public GitHub.

Compact deletes low value lines from long transcripts verbatim, cutting them by half or more, Reflex returns typed decisions as probabilities, and the Router picks a model from OpenAI, Anthropic, Google, DeepSeek, Z.ai or Moonshot for each prompt. Morph also hosts open weight models such as Kimi K3 and DeepSeek V4 behind the same API, and sells dedicated inference on B200 GPUs.

The pieces reach agents through an API that speaks both the OpenAI and Anthropic formats, a TypeScript SDK with adapters for the Anthropic, OpenAI and Vercel AI SDKs, a command line tool, six installable skills for Claude Code, Codex, Cursor and OpenCode, and a local MCP server exposing edit and search tools. Pricing is per token, with 200 free requests a month, pay as you go top ups from $10, a $200 a month Scale plan carrying $400 of credits, and custom Enterprise terms with SSO. Morph holds a SOC 2 Type I report, deletes content within 30 days on standard plans, offers zero data retention to enterprise customers, and can run self hosted, on premises or air gapped.

It fits teams building their own coding agents, and platforms such as app builders, that want faster edits and cheaper search without training models themselves; named customers include JetBrains, Vercel, Webflow, Warp, Binance and Block. It fits less well for a team that wants a finished coding agent, since every component waits for another agent to call it, and for buyers who need a SOC 2 Type II report or a named data residency region today.

Vendor details

Canonical URL

https://www.morphllm.com

Subcategory

Fast specialized models and a code search subagent that run inside other companies' coding agents

Funding status

Private company. AutoInfra, Inc. of San Francisco, doing business as Morph, part of the Y Combinator Summer 2023 batch.

Company status

independent

Use cases & customers

Primary use cases

merging coding agent edits into full files at high speedcode search as a subagent inside a coding agentcompacting long agent transcripts to cut context costrouting each prompt to a cheaper or stronger modelhosting open weight models for coding agents behind one API

Target customers

companies building their own coding agentsapp builder and vibe coding platformsdevelopers running Claude Code, Codex, Cursor or OpenCode on large codebasesAI platform teams cutting model cost inside agent loops

Deployment options

SaaS APIDedicated inferenceSelf hosted or on premises, including air gapped

Integrations

Adapters for the Anthropic, OpenAI and Vercel AI SDKs, with setup guides for Claude Code, Codex, Cursor, OpenCode, Windsurf, Amp and Antigravity, and gateway routes through Claude Code Router and LiteLLM. Fast Apply writes into E2B, Modal, Daytona, Vercel Sandbox and Cloudflare sandboxes. Morph's models are also offered through OpenRouter.

In practice

An app builder's agent writes changes as short snippets with placeholder comments. Fast Apply merges each snippet into the full file in under a second, so users watch their app update almost as soon as the agent finishes thinking.

A team running Claude Code on a large monorepo installs the WarpGrep skill. Searches that used to fill the main context with grep output now return only the file and line ranges that matter, and long sessions stop hitting the context limit.

A coding agent platform sends every prompt through Morph's Router, which sends simple refactors to a small open weight model and hard debugging work to a frontier model, cutting the monthly model bill without changing the agent's code.

Sources & related URLs

Related / legacy domains

https://docs.morphllm.com https://status.morphllm.com https://www.morphllm.com/mcp

Research sources

https://www.morphllm.com/ https://www.morphllm.com/pricing https://www.morphllm.com/products/fastapply https://www.morphllm.com/products/warpgrep https://www.morphllm.com/products/models https://www.morphllm.com/benchmarks https://www.morphllm.com/privacy https://www.morphllm.com/privacy/tos https://docs.morphllm.com/endpoints https://docs.morphllm.com/sdk/components/warp-grep/index https://docs.morphllm.com/skills https://docs.morphllm.com/dedicated-endpoints https://trust.delve.co/morph

Agentic Index coverage score

7.5 / 14 capabilities · 54%

Integrations & Tool Calling	Partial
Morph hands coding agents two tools, an edit tool backed by Fast Apply and a codebase search tool backed by WarpGrep, with ready adapters for the Anthropic, OpenAI and Vercel AI SDKs and setup guides for Claude Code, Codex, Cursor, OpenCode, Windsurf, Amp and Antigravity. Fast Apply writes merged files into E2B, Modal, Daytona, Vercel Sandbox and Cloudflare sandboxes. Beyond the repository and public GitHub, its components connect to no business systems, and there is no connector catalog or framework for adding other tools. Sourcedocs.morphllm.com/sdk/components/warp-grep/toolread 2026-11-08
Workflow Orchestration	Partial
WarpGrep runs as a subagent: the main agent hands it a search, and it works through its own loop of grep, file reads and directory listings in a separate context window before returning file and line spans. The Router chooses a model for each prompt, and Reflex returns a scored decision the calling agent can branch on. Planning, sequencing and handoffs between the customer's agents stay in the customer's own code, with Morph supplying individual steps inside that loop. Sourcedocs.morphllm.com/sdk/components/warp-grep/indexread 2026-11-08
Knowledge Grounding & RAG	Partial
WarpGrep finds context at the moment it is needed. It searches a local repository live, searches public GitHub in Morph's cloud, and the command line tool keeps shallow clones of GitHub repositories in a local cache. An embeddings model with 1,024 dimensions and a rerank model are available for teams that build their own retrieval, with the vector store left to the customer. Morph keeps no persistent index of a customer's code or documents. Sourcedocs.morphllm.com/sdk/components/warp-grep/indexread 2026-11-08
Human Oversight & Guardrails	Not documented
Edits, search results and decisions go back to the calling agent, and that agent's harness decides whether a person reviews them. The command line tool can preview an edit with a dry run before writing it, Reflex lets a team set its own confidence threshold for automatic actions, and admins can set monthly budgets per member that pause that member's keys. Morph has no approval step or review queue of its own. Sourcedocs.morphllm.com/cliread 2026-11-08
Security, Identity & Governance	Full
Organizations have admin and member roles, organization and personal API keys, and monthly budgets per member that admins set and that pause the member's keys when reached, with SSO on Enterprise. Morph holds a SOC 2 Type I report, shown on its trust center. Content on free and paid plans is deleted within 30 days, zero data retention is available to enterprise customers under a signed addendum and is the default on dedicated inference, and customer data is never used for training. Sourcetrust.delve.co/morphread 2026-11-08
Observability & Auditability	Partial
Dedicated endpoints come with a dashboard of requests, latency, cache hits, tokens and status, and Fast Apply carries a usage dashboard and a log of merges, diffs and errors. The command line tool logs metadata only, and a public status page covers the API. These views report volume and outcome for each call; they do not trace the steps an agent took across a session or show why it made each tool call. Sourcedocs.morphllm.com/dedicated-endpointsread 2026-11-08
Memory & State Persistence	Not documented
No memory is kept for the agents that call Morph. A prefix cache can hold a prompt prefix for up to 24 hours to cut cost and latency, and a run identifier keeps a model's working cache warm until it has been idle for 15 minutes. Compact shrinks long transcripts so the calling agent can store and reuse them, but the storing happens in the customer's own system, and no store carries facts from one session to the next. Sourcedocs.morphllm.com/sdk/components/cachingread 2026-11-08
Deployment & Data Residency	Full
Deployment runs in Morph's cloud or the customer's: Fast Apply can be self hosted, including air gapped, and on premises and private cloud deployments are sold through the sales team. Dedicated inference runs on B200 or B300 GPUs. Standard processing happens in the United States, with regional data centers for paid and enterprise customers and no region picker in the product. Sourcemorphllm.com/products/fastapplyread 2026-11-08
Prebuilt Agents / Templates / Packs	Full
Six installable skills package Morph for coding agents: explore, feature research, code research, fast apply, the Morph command line tool and Morph migrate, each loading into Claude Code, Codex, Cursor or OpenCode. Fast Apply, WarpGrep, Compact, Reflex and the Router are each sold as a ready model under its own model ID, and an examples repository on GitHub carries working integrations for agent frameworks and sandboxes. Sourcedocs.morphllm.com/skillsread 2026-11-08
Triggers & Channel Coverage	Not documented
Every call starts in the customer's code or agent. Requests arrive over the REST API, the TypeScript SDK, the command line tool that agents run in their shell, installed skills, or the local MCP server, and credits can be topped up through the API. Nothing starts work on its own: the platform has no event subscriptions, schedules, webhooks or inbound queues, and each component returns its result to whatever called it. Sourcedocs.morphllm.com/endpointsread 2026-11-08
Model Flexibility & Routing	Full
Customers pick the model by ID on every call. The catalog covers Morph's own models for apply, search, compaction, decisions and routing, and hosted open weight models including Kimi K3 and K3.1 Pro, GLM 5.3 and its Flash version, DeepSeek V4 Flash variants, Qwen 3.6 and MiniMax M3. The Router recommends a model from OpenAI, Anthropic, Google, DeepSeek, Z.ai or Moonshot for each prompt, and the customer makes the call with it. Sourcemorphllm.com/products/modelsread 2026-11-08
APIs / SDKs / MCP Extensibility	Full
The API at api.morphllm.com speaks the OpenAI chat completions format and the Anthropic messages format, with further routes for compaction, routing, embeddings and rerank, and authenticates with a bearer token or an API key header. A TypeScript SDK, with an edge build, wraps every component, and Python works through the OpenAI client. A command line tool and a local MCP server exposing edit and WarpGrep search tools round out the developer surface. Sourcedocs.morphllm.com/endpointsread 2026-11-08
Testing, Debugging & Optimization	Partial
Benchmarks of Morph's own components sit on its site, among them WarpGrep reaching 0.73 F1 in 3.8 steps and cutting cost by 15 percent and time by 19 percent on a public software engineering benchmark, and an open coding agent benchmark on GitHub lets a team rerun comparisons with and without Morph's models. The platform has no product surface for scoring a customer's own agent on saved test cases or for gating a release on the results. Sourcemorphllm.com/benchmarksread 2026-11-08
Browser / Computer-use	Not documented
All of the components work on text: code, transcripts and prompts passed through the API. None of them drives a browser, a desktop or a web interface, and no hosted browser is offered. WarpGrep's tools are limited to grep, file reads and directory listings inside a repository, and Reflex accepts images only as input when it scores a decision for the calling agent. Sourcedocs.morphllm.com/models/reflexread 2026-11-08

The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded

Pricing

Pay as you go from a $10 top up, billed per token, with 200 free requests a month. Scale is $200 a month for $400 of credits. Enterprise is custom.

Per million input, cached and output tokens for each model, per request for the Router, drawn from a prepaid credit balance, with dedicated inference billed per GPU hour.

Free tier

Included quota

The free tier covers 200 requests a month. Scale includes 40 million credits each month, worth $400 at list rates, for $200.

What is public

Per token rates for Morph's own models and the hosted open weight models, the free allowance, the pay as you go minimum, the Scale plan, the credit fee and the startup credit offer.

Billing mechanics

Customers prepay credits by card, or by API in USDC, and each call draws them down at the model's per token rate. Enterprise and dedicated inference are contracted through sales.

Cost watchouts

The 5.5 percent credit fee applies to every top up. Hosted frontier class open models such as Kimi K3 cost far more per token than Morph's own small models, at $14.00 per million output tokens. Dedicated inference carries a 90 day initial term.

Variable cost rationale

Every component bills per token or per request, so cost follows how many edits, searches and compactions a coding agent runs and how large each one is. The Scale plan converts a flat $200 into $400 of credits, and dedicated inference moves to per GPU hour billing on a 90 day initial term.

Additional watchouts

The WarpGrep price appears three different ways across Morph's pages, so a quote is the reliable figure. Standard plans delete content within 30 days; zero data retention needs an enterprise addendum or dedicated inference.

Overage / add-ons

Usage draws down the credit balance; more credits are bought by top up, with a 5.5 percent fee and a $0.80 minimum on each purchase.

Sales call required

Mixed (some tiers require a call)

Free / trial

200 free requests every month, $5 added to the first top up, and up to $1,000 in credits for startups.

Lowest paid plan

Pay as you go from a $10 top up, with $5 added to the first one; Scale at $200 a month for 40 million credits worth $400

Commercial notes

Morph is sold by AutoInfra, Inc. of San Francisco, a Y Combinator Summer 2023 company. Named customers include JetBrains, Vercel, Webflow, Warp, Binance and Block.

Key ambiguities

The free tier is given as 200 requests a month on the pricing page and as $10 a month of compute on the WarpGrep page, and the WarpGrep rate differs between the pricing page, the docs and the product page.

Cancellation / refund

Credits are prepaid. Dedicated inference starts on a 90 day initial term.

Support SLA / resale

Enterprise adds priority support and custom rate limits.

Missing data

A settled WarpGrep rate, dedicated inference rates per GPU hour, enterprise pricing, and prices for the embeddings and rerank models.

Agentic Index verified 2026-11-08

Official pricing page Docs

Data confidence: medium